This article explores the distribution of morphological case in Spanish nominalizations and shows that there is a connection between morphological case and event structure. Most nominals govern genitive case de on their internal arguments but some allow or require a different morpheme, a, reminiscent of differential object marking. I argue that the event structure of the nominalization is the crucial factor in choice of a, inasmuch as a is limited to process nominals that do not entail a change of state. The same distinction between process and change-of-state nominals is then extended to two other empirical puzzles regarding the interpretation of genitive arguments in nominalizations. I present a formal analysis assuming a syntax of events inspired by Ramchand (2008).
Spanish nominalizations generally require genitive case on their arguments, as is common crosslinguistically; this is shown in (1a). However, some nominalizations allow or require that their internal arguments be introduced by a, as shown in (1b). (1c) shows that there is a clear division between nominalizations that allow a and those that do not.1
La captura de Juan por el perro fue sorprendente.
the capture GEN Juan by the dog was surprising
‘The dog’s capture of Juan was surprising.’
El ataque del perro a Juan fue sorprendente.
the attack GEN.DEF dog DOM Juan was surprising
‘The dog’s attack on Juan was surprising.’
*La captura del perro a Juan fue sorprendente.
the capture GEN.DEF dog DOM Juan was surprising
Among the many nominals that behave like ataque are golpe ‘hit’, miedo ‘fear’, and acusación ‘accusation’. Nominals that behave like captura include destrucción ‘destruction’, limpieza ‘cleaning’, retraso ‘delay’, and others (a more complete list is in section 3.1).
I will refer to this phenomenon as n-DOM, in part because of its superficial similarity to the phenomenon of differential object marking (DOM) in the verbal domain (henceforth v-DOM), and in part because to the extent that the literature has mentioned n-DOM it has done so in the context of an analysis of v-DOM (as in Torrego 1998). I exemplify v-DOM in (2).
El perro atacó a Juan.
the dog attacked DOM Juan
El perro capturó a Juan.
the dog captured DOM Juan
Spanish v-DOM is a well-known phenomenon with an extensive literature, mostly focused on trying to account for its distribution (see Fábregas 2013b for a detailed overview). In contrast, n-DOM is unexplored territory: I am only aware of a two-paragraph discussion in Torrego 1998, and the comprehensive grammar of Spanish compiled by Bosque and Demonte (1999) does not mention n-DOM in any of its more than 5,000 pages.
In two respects, n-DOM resembles v-DOM: n-DOM is found in argument-taking nominals, and it affects exclusively the internal argument. Phonologically, n-DOM and v-DOM are also identical. However, the parallels stop there. The contrast between (1c) and (2b) already points to a fundamental difference between v-DOM and n-DOM: although the verb capturar is compatible with v-DOM, the noun captura is not.
Notice that the distribution of genitive case in examples (1) and (2) poses another puzzle. Since Picallo 1991, it has commonly been assumed that in event nominalizations the internal argument but not the external argument is a true argument of the nominal (see in particular Alexiadou’s (2001) extensive discussion of a database that includes several languages). Evidence for this claim can be seen in the case distribution of (1a): the external argument appears in an adjunct por/by-phrase while genitive de/of is reserved for the internal argument. The claim is certainly true of a subset of nominals, those that look like captura, as shown in (3a). But it does not hold of ataque nominals. In the latter type of nominal, the external argument may appear in genitive case, as shown in (3b–c). Moreover, the genitive constituent bearing the external θ-role seems to be a bona fide noun phrase argument; a simple test like reflexive binding shows that it is not an adjunct PP, as shown in (3c).
la captura del perro
the capture GEN.DEF dog
‘the dog’s capture’ (the dog is patient)
el ataque del perro
the attack GEN.DEF dog
‘the dog’s attack’ (the dog is agent)
el ataque de Juan a sí mismo
the attack GEN Juan DOM himself
‘Juan’s attack on himself’
Finally, the phenomena exemplified in (1) and (3) point to a third puzzle. Anderson (1977) argues that only some internal arguments can appear in the Saxon genitive construction, those that she defines as being affected (see also the discussion in Grimshaw 1990, Alexiadou 2001, Smirnova 2015). Consider the examples in (4). In John’s capture, John must be the internal argument, but in John’s attack, John is the external argument. Spanish does not have Saxon genitives but it does have possessive adjectives, as shown in (4c–d). Let’s use the term s-gen as a cover term for the English Saxon genitive and for English and Spanish possessive adjectives. Interestingly, the interpretation of su in (4c–d) parallels the interpretation of John’s/his in (4a–b): it is a theme in (4c) and an agent in (4d).
John’s / his capture
John’s / his attack
‘her / his capture’
‘her / his attack’
I hypothesize that these three properties do not converge on the same class of nouns by chance. That is, it is not by chance that the ‘attack’ nouns have the properties of assigning n-DOM, having a de-genitive external argument in Spanish, and having an s-gen external argument in Spanish and English, while the ‘capture’ nouns have the opposite properties. Further, I argue that what underlies these grammatical properties is the structure of the event that the nouns denote. Thus, this article is meant to be a contribution to our understanding of the role of inner aspect in shaping the grammatical structure of a predicate.
The contrast between (1b) and (1c) is the central datum of this article. I argue that n-DOM is not possible when the event denoted by the noun entails a change of state that involves the internal argument; correspondingly, n-DOM is possible when the event does not entail a change of state. I account for this by means of a syntax of events directly inspired by Ramchand (2008). As I will show, this syntax of events also accounts for the second and third puzzles. In the course of the discussion, I show that Spanish n-DOM and v-DOM are distinct phenomena. In particular, n-DOM cannot be equated with structural accusative or inherent case.
The article is structured as follows. Section 2 presents a descriptive grammar of n-DOM in Spanish and motivates the need to study it as an independent linguistic phenomenon. This section goes into the features of n-DOM at some length because, as mentioned above, there is no previous literature to refer to. Section 3 presents my account of the puzzle exemplified in (1) and (2), and section 4 discusses the second and third puzzles. Finally, section 5 takes up an additional issue: there are some nominals for which n-DOM is obligatory and others for which it is optional. Section 6 concludes.
2 A Descriptive Grammar of n-DOM
This section is organized as follows. Section 2.1 provides a quick refresher on the properties of v-DOM in Spanish while section 2.2 shows that n-DOM has none of the properties that define v-DOM. Along the same lines, section 2.3 shows that n-DOM cannot be equated with structural accusative case and section 2.4 shows that it is unlikely to be an instance of inherent case. Section 2.5 discusses adicity and summarizes this part of the discussion.
2.1 Conditions on v-DOM
Juan golpeó a / *Ø Carlos.
Juan hit DOM Carlos
‘Juan hit Carlos.’
Juan golpeó *a la instalación.
Juan hit DOM the installation
‘Juan hit the installation.’
The second factor that defines the appearance of v-DOM is Silverstein’s (1976) referentiality scale, successfully deployed by Aissen (2003), as well as many others, to analyze v-DOM. Spanish v-DOM ranks quite low in the referentiality scale: v-DOM is obligatory with pronouns, definite and specific DPs, proper names, and strong quantifiers; it is optional with nonspecific objects. This is exemplified in (6).
La policía está persiguiendo a una gestora corrupta.
the police is pursuing DOM a manager corrupt
‘The police are pursuing a (specific or nonspecific) corrupt manager.’
La policía está persiguiendo una gestora corrupta.
‘The police are pursuing a (nonspecific) corrupt manager.’
Although v-DOM is sometimes referred to as a “prepositional direct object,” the a is not a preposition. According to traditional tests, the phrase headed by v-DOM is an ordinary noun phrase direct object; for instance, it is promoted to subject in a passive sentence. Instead, the a should be regarded as a spell-out of accusative case.
2.2 Conditions on n-DOM
As mentioned, n-DOM and v-DOM have one property in common: they show up on internal arguments. Other than that, the conditions on n-DOM do not reproduce the conditions on v-DOM. First, there is no animacy constraint on n-DOM, as the examples in (7) show.
el golpe a la instalación / a Carlos
the hit DOM the installation / DOM Carlos
‘the hitting of the installation / of Carlos’
el ataque a la ciudad / a la candidata
the attack DOM the city / DOM the candidate
‘the attack on the city / on the candidate’
el miedo a las tormentas / a la condesa
the fear DOM the storms / DOM the countess
‘the fear of storms / of the countess’
Second, there is no specificity effect with n-DOM. In (8), for example, the nominalization persecución ‘persecution, pursuit’ can assign n-DOM. I use the indicative vs. conditional contrast in the main clause as well as the indicative vs. subjunctive contrast in the relative clause in order to bring out the specific and nonspecific readings in the DP ‘a manager that is honest’. The idea is that if the matrix verb is perfect and the subordinate verb is indicative, the specific reading is obligatory. If the matrix verb is conditional and the subordinate verb is subjunctive, the nonspecific reading becomes prominent. These changes make no difference: either the genitive marker or the DOM marker is possible in each case.
María presenció / presenciaría sin remordimiento . . .
Maria witnessed / would.witness without remorse
. . . la persecución de / a una gestora que fue / fuera honrada.
the persecution GEN / DOM a manager that was.INDIC / is.SUBJ honest
‘. . . the persecution of a manager that was / is honest.’
Thus, I conclude that the conditions on the appearance of n-DOM do not match those on the appearance of v-DOM.
2.3 n-DOM Is Not Structural Accusative Case
Two independent lines of research developed in the last decade would seem to conspire to suggest an analysis of n-DOM as an expression of structural accusative case. The first line of research argues that in many languages, v-DOM is indeed a spell-out of structural accusative case assigned by little v or a functional category within the v-VP/√P complex (e.g., Rodríguez-Mondoñedo 2007, López 2012). The evidence in favor of this analysis seems fairly conclusive. For instance, in passive sentences, arguments bearing v-DOM assume the grammatical function of subject; additionally, when v-DOM arguments are pronominalized, the pronoun exhibits overt accusative morphology.2
The second line of research involves nominalizations. In recent years, an analysis of nominalizations has become mainstream in which the noun phrase dominates a full-fledged verb phrase (see Alexiadou 2001, Embick 2010, Borer 2012).
Mainstream analysis of nominalizations
[nP n [vP v [√]]]
The ability of nominalizations to denote complex events (in the sense of Grimshaw 1990) is attributed to the presence of the vP, which is in charge of introducing the event variable and the arguments of the event variable. It follows that a noun phrase may denote a complex event only to the extent that it dominates a vP.
Putting these two lines of research together, one could conclude that n-DOM is nothing but an instance of v-DOM that has been swallowed by a nominal structure. This seems to be the state of affairs that obtains in Hebrew, as argued by Borer (2013b); Hebrew is a language in which at least some nominalizations clearly seem to include a vP with an internal argument that can bear n-DOM (= v-DOM). However, Spanish is not like Hebrew, and n-DOM cannot be regarded as a spell-out of structural accusative case.
There are at least two reasons why n-DOM cannot be equated with accusative case. First, v-DOM alternates with a DP that does not bear an external form of case.
Koch cazó unos conejos.
Koch caught some rabbits
Marine cazó a un hombre.
Marine caught DOM a man
n-DOM instead alternates with genitive. Notice in particular that a DP without a case mark is not grammatical within a nominalization.
la caza al hombre escapado
the hunting DOM.DEF man escaped
‘the hunting of the escaped man’
la caza de los conejos
the hunting GEN the rabbits
‘the hunting of the rabbits’
*la caza los conejos
Second, n-DOM is not a dependent case (on this notion, see Marantz 1991, Baker 2015). Accusative case is dependent on another argument receiving nominative case. Example (12) shows that v-DOM, which is a spell-out of accusative case, is not possible in passive sentences.
*Fue atacado a Juan por el perro.
was attacked DOM Juan by the dog
(12) is ungrammatical because in the absence of an external argument, the internal argument must assume the role of subject and become a nominative DP. (12) contrasts with (13a–b), examples with an event nominal. (13a) shows that n-DOM is possible when the external argument is introduced with a by-phrase, adopting a shape that is strongly reminiscent of a passive structure. (13b) shows that the external argument can be absent altogether. n-DOM is not dependent on any other case within the nominal. If n-DOM were a dependent case, the DPs in (13) should be ungrammatical.
el ataque a Juan por el perro
the attack DOM Juan by the dog
‘the dog’s attack on Juan’
el ataque a la ciudad
the attack DOM the city
‘the attack on the city’
Moreover, there are reasons to doubt that (9) is indeed the best analysis of event-denoting nominals in Spanish. As I argue in López 2015, there are complex event nominals in Spanish that can’t possibly be built on verbs. (14) and (15) exemplify complex event nominals that do not include a verbal base: the roots √BASTON- and √MIED- cannot be used as bases to form verbs. For instance, the noun bastonazo is built on the noun bastón ‘stick, club’; the suffix azo by itself brings in the event meaning.
el bastonazo del policía al manifestante
the “bastonazo” GEN.DEF police.officer DOM.DEF demonstrator
‘the police officer’s hitting the demonstrator (with a stick)’
el miedo de Juan a las arañas
the fear GEN Juan DOM the spiders
‘Juan’s fear of spiders’
Another reason to doubt that Spanish nominalizations include a vP is that they do not license adverbs, as shown in (16).
la injusta acusación al sargento
the unjust accusation DOM.DEF sergeant
‘the unjust accusation against the sergeant’
*la injustamente acusación al sargento
the unjustly accusation DOM.DEF sergeant
We can conclude, then, that Spanish nominalizations (except those derived from infinitives) do not include a full-fledged vP. Mainstream analyses of complex words generate them uniformly as head-complement structures, starting with the root and moving upward as in (17b). In López 2015, I argue that it is also possible for affixes to merge in a separate workspace, building an output that subsequently merges with a lexical root as in (17a).
acusación: [[acus]√ [[a]v cion]]nLópez 2015
acusación: [[[acus]√ a]v cion]n Mainstream analysis
In (17a), the verbal morpheme is embedded within the syntactic head n and not able to project onto a phrase. Presumably, within this structure v should not be able to license n-DOM either.
2.4 n-DOM Is Not Inherent Case
Torrego (1998) argues that some instances of v-DOM in Spanish involve inherent case: namely, instances containing verbs for which v-DOM on the object seems to be obligatory. Torrego’s list includes the verbs acusar ‘accuse’, castigar ‘punish’, ofender ‘offend’, empujar ‘push’, golpear ‘hit’. In addition, Torrego claims that these instances of inherent accusative are the ones that survive in nominalizations. Consequently, n-DOM is in fact, according to Torrego, inherent accusative case. Indeed, it is true that the nominalizations that correspond to the verbs in Torrego’s list all exhibit n-DOM.
La acusación al sargento fue instruida por el fiscal de la audiencia
the accusation DOM.DEF sergeant was filed by the attorney of the audience
‘The accusation against the sergeant was filed by the district attorney.’
However, there are many examples that do not follow Torrego’s generalization. As mentioned in section 2.3, there are even examples of n-DOM in nominalizations that do not derive from a verbal base (see (13)–(15)).
Here, I further argue that conceiving of n-DOM as an inherent accusative case is highly implausible. This is the structure of the argument: In section 2.3, I showed that n-DOM is not structural accusative case. In this section, I argue that inherent dative case is not possible in nominals. The two conclusions together strongly suggest that n-DOM cannot be a form of inherent accusative case.
Ditransitive predicates in Spanish come in two forms, exemplified in (19a–b).
María le entregó el paquete a Susana.
Maria CL delivered the package DAT Susana
‘Maria delivered the package to Susana.’
María entregó el paquete a Susana.
Maria delivered the package to Susana
la entrega del paquete a Susana
the delivery GEN.DEF package to Susana
‘the delivery of the package to Susana’
In the first form, exemplified in (19a), the indirect object is doubled by a dative clitic, with the result that the semantic and syntactic properties of the construction make it akin to the English double object construction (see Demonte 1995, Bleam 1999, Cuervo 2003). The a-morpheme that introduces the indirect object should be regarded as a dative case marker.3 In the second form—that is, when the indirect object is not doubled by a clitic—the indirect object is a goal, the equivalent of the English indirect object with to. In this case, a is a preposition, as shown in the gloss of (19b).
Finally, (19c) exemplifies a nominalization built on the same root, √ENTREG-. The question is whether the a in the nominal is the dative case marker or the preposition. The gloss shows that I have decided it is the preposition. Examples (20a–d) show how I reached this conclusion.
Juan le construyó una casa a su padre.
Juan CL built a house DAT his father
‘Juan built his father a house.’
Juan construyó una casa para su padre.
Juan built a house for his father
la construcción de la casa para su padre
the construction GEN the house for his father
‘the construction of the house for his father’
*la construcción de la casa a su padre
the construction GEN the house DAT his father
(20a) exemplifies a Spanish applicative construction. The dative clitic acts as the applicative morpheme and the indirect object is a beneficiary introduced by the dative case marker a. In (20b), without a dative clitic, the beneficiary is introduced by the preposition para ‘for’. As examples (20c–d) show, nominalizations only admit the preposition para, not the dative case marker a. This suggests that dative case is unavailable in nominalizations.
A similar conclusion results from considering psychological (psych) predicates. Some psych verbs select dative case on the experiencer, as shown in (21a). However, the corresponding nominalizations are incompatible with dative case, as shown in (21b–c).
A María le preocupa la salud de su madre.
DAT Maria CL worry the health of her mother
‘Maria is worried about her mother’s health.’
*la preocupación a María
the concern DAT Maria
la preocupación de María
the concern GEN Maria
Examples (20) and (21) teach us that inherent dative case is unavailable in nominals. Since structural accusative case is also unavailable (section 2.3), I conclude that the availability of inherent accusative case in nominals is highly improbable. (See also Alexiadou’s (2001:44) claim that there are no nominalizations based on double object constructions, and Pujalte’s (2009) observation that nominalizations do not accept applicative benefactives.)
el trabajo eficiente de María
the work efficient GEN Maria
‘Maria’s efficient work’
*el trabajo eficiente a María
the work efficient DOM Maria
la llegada tardía de Juan
the arrival late GEN Juan
‘Juan’s late arrival’
*la llegada tardía a Juan
the arrival late DOM Juan
Since n-DOM only affects internal arguments, the lack of n-DOM in unergative predicates seems to follow directly. In the case of unaccusatives, the account is more complex. The absence of accusative case in verbal unaccusative predicates follows from Burzio’s Generalization or from the assumption that accusative case is a dependent case, as argued by Marantz (1991) and Baker (2015). Along these lines, one could hypothesize that n-DOM is a form of dependent case to account for the ungrammaticality of (23b). However, as I showed in section 2.3, n-DOM is not a dependent case. Consequently, the ungrammaticality of (23b) remains a puzzle at this point (I take up the question again and provide an account in section 3.1).
Finally, the theme argument of ditransitives is also sharply ungrammatical in n-DOM forms. Only the genitive marker is possible. As (24a–b) show, the theme argument may appear in genitive case, and as (24c) shows, it cannot exhibit n-DOM.
la entrega de un paquete a María por Juan
the delivery GEN a package to Maria by Juan
‘the delivery of a package to Maria by Juan’
?la entrega de Juan de un paquete a María
the delivery GEN Juan GEN a package to Maria
*la entrega a un paquete a María
the delivery DOM a package to Maria
The Distinctness Condition (henceforth DC) provides a temptingly straightforward account of the ungrammaticality of (26c); however, the temptation must be resisted. Versions of the DC have been used by grammarians working on Spanish to account for the impossibility of v-DOM in (25).
María presentó *a / Ø Susana a su padre.
Maria introduced DOM Susana DAT her father
‘Maria introduced Susana to her father.’
The idea is simple: the DC is a restriction that prevents two arguments headed by the case marker a to be found in the same VP. Richards (2010) develops a theory of the DC and expands it to several empirical realms, including the datum in (25). His theory can be summarized as follows:
Distinctness (* 〈α,α〉)
If a linearization statement 〈α,α〉 is generated, the derivation crashes.
In Richards’s view, linearization takes place by phase. Thus, (26) prevents two arguments headed by the same case morphology from being linearized in the same phase. One could then extend the analysis to (24c), where there are also two arguments introduced by a. However, a more detailed consideration of the data shows that the ungrammaticality of (24c) is unrelated to that in (25) and cannot be due to the DC. As the DC predicts, v-DOM in (25) becomes grammatical—in fact, obligatory—when the indirect object is omitted or when the indirect object is only a clitic pronoun. In (27a–b), a Laura can only be interpreted as the direct object.
Les presento a Laura.
CL.2.DAT introduce.1SGDOM Laura
‘I introduce Laura to you.’
(In the midst of a loud ovation, Maria went on stage in order to . . .)
. . . presentar a Laura.
‘. . . introduce Laura.’
Thus, we can use absence of an explicit indirect object to test whether the ungrammaticality of (24c) is due to the DC: if it is, the judgment should improve when the indirect object is not explicit. This expectation is not fulfilled: (24c) remains ungrammatical when the indirect object is omitted, as shown in (28).
*la entrega al paquete
the delivery DOM.DEF package
In (29), with the nominalization presentación ‘introduction’ built on the verb presentar ‘introduce’, the presence of a forces an indirect object reading, which indicates that a is the dative preposition. The direct object reading is only possible with the genitive de.
la presentación a María
the introduction to Maria
la presentación de María
the introduction GEN Maria
Thus, the DC does not provide an account of the ungrammaticality of (24c).
In this section, we have learned that n-DOM cannot be equated with structural accusative or inherent case; moreover, it is not subject to the same restrictions that apply to v-DOM. Thus, we must conclude that n-DOM is an independent linguistic phenomenon. In terms of distribution, n-DOM displays three puzzling empirical facts that I propose to account for in the following section: (a) the contrast between ataque ‘attack’ and captura ‘capture’ with respect to the availability of n-DOM, (b) the absence of n-DOM with unaccusatives, and (c) the absence of n-DOM in ditransitives.
3 Event Structure and n-DOM
3.1 Process and Change-of-State Nominals
I propose that the distribution of n-DOM can be accounted for by appealing to the subevent structure of the nominal. Let’s assume that we can divide events into two types (for analyses of event structure, see Pustejovsky 1991, Croft 1993, Levin 1999, Ramchand 2008, among many others; see also Levin and Rappaport Hovav 2005 for an overview). The particular division of event structures that I use here has a clear precedent in Levin 1999.
Change of state
A change-of-state event includes a state S1 that acts as a source or initiation point. Call S1 the initial state. S1 is subject to a process that acts on the internal argument, and the output of the process is a state S2 in which some property of the internal argument has been altered. The alteration can involve
creation (the internal argument denotes something that exists in S2 but did not in S1: build, breed, grow, develop )
modification or destruction (the internal argument denotes something that has or lacks a property in S2 that it lacked or had in S1: kill, mutilate, dissolve, colonize, break, open)
displacement (the internal argument denotes something that occupies a different location in S2 than in S1: transport, bury, lift, lower)
A process event includes a state S1 that acts as a source or initiation point. S1 is subject to a process that acts on the internal argument, but the output of the process does not entail any transformation of the internal argument. Examples: push, attack, pursue, help. Compare mutilate (change of state) with torture (process), capture (change of state) with hunt (process), destroy (change of state) with attack (process).
In light of this distinction, consider now the lists of nouns in (30) and (31). The nouns in (30) accept n-DOM, while the nouns in (31) do not (for examples, see (32) and (33)). All of the predicates in (30) and (31) are transitive. They are listed in alphabetical order according to the nominalizing morpheme or word marker—a presentation intended to convince the reader that almost any nominal suffix in Spanish is compatible with n-DOM. Some of the affixes listed attach directly to the root as word markers ([a], [e], [o], [on]). The other affixes attach to a root that bears the verbal theme vowel and are more properly regarded as nominalizations. As pointed out in footnote 1, Spanish nominals derived directly from the root can take arguments and be complex event nominals (in the sense of Grimshaw 1990), and therefore I make no distinction between root-derived and verb-derived event-denoting nominals.
The classification of nouns into the n-DOM and no-n-DOM categories is based on a survey of eight native speakers of Peninsular Spanish, who were asked to provide acceptability judgments on sentences that included n-DOM. There was individual variability because some speakers were a little more restrictive in their acceptance of n-DOM, and these data only present the judgments of a sizable majority—if a nominalization is listed in (30), at least six of the eight consultants accepted it. I make no claim that the pattern revealed by this survey can be extended to other varieties of Spanish. In constructing the stimuli, I made sure that speakers were not presented with spurious examples based on light verb constructions (as in ‘give a push to someone’) in which the apparent n-DOM might in fact be a homophonous indirect object of the light verb.
[a]: estafa ‘fraud, swindle’, caza ‘hunt’, crítica ‘criticism’
[ada]: puñalada ‘stabbing’
[azo]: bastonazo ‘blow’, puñetazo ‘punch’
[aje]: chantaje ‘extortion’
[cion]: persecución ‘persecution, pursuit’, inspección ‘inspection’, circunnavegación ‘circumnavigation’
[da/do]: despedida ‘farewell’, cogida ‘catching (as in a bullfight)’, bofetada ‘slap’,- patada ‘kick’, barrida ‘sweep’
[e]: ataque ‘attack’, combate ‘combat’, golpe ‘blow’
[miento]: reconocimiento ‘recognition, acknowledgment’, acompañamiento ‘accompaniment’, acatamiento ‘compliance’, seguimiento ‘follow-up’
[ncia]: advertencia ‘warning’, obediencia ‘obedience’
[nza]: alabanza ‘praise’, venganza ‘revenge’, esperanza ‘hope’, añoranza ‘homesickness’
[o]: beso ‘kiss’, abrazo ‘hug’, acoso ‘harassment’, consejo ‘advice’, repaso ‘revision’
[on]: empujón ‘push’, achuchón ‘cuddle’
[ura]: tortura ‘torture’
(31) No n-DOM
[a]: entrega ‘delivery’, mejora ‘improvement’
[aje]: fichaje ‘hiring’, maquillaje ‘makeup’
[cion]: colonización ‘colonization’, cristianización ‘Christianization’, catalanización ‘Catalanization’, mutilación ‘mutilation’, disolución ‘dissolution’
[da/do]: bordado ‘embroidering’, barnizado ‘varnishing’
[e]: trueque ‘exchange’, ligue ‘hookup’, transporte ‘transport’
[miento]: encubrimiento ‘cover-up’, descubrimiento ‘discovery’
[ncia]: transferencia ‘transfer’
[nza]: crianza ‘breeding, growing’, matanza ‘slaughter’
[o]: despido ‘firing’, entierro ‘burial’, retraso ‘delay’, traslado ‘transfer’
[ura]: captura ‘capture’, rotura ‘break’
(32) Some n-DOM examples
el abrazo a mis sobrinos
the hug DOM my nephews
‘the hug of my nephews’
la caza a los politicos corruptos
the chase DOM the politicians corrupt
‘the chasing of corrupt politicians’
la crítica a las ideas de Chomsky
the criticism DOM the ideas GEN Chomsky
‘the criticism of Chomsky’s ideas’
el reconocimiento a sus valerosas acciones
the recognition DOM her/his courageous deeds
‘the recognition of her / his courageous deeds’
la advertencia a mis hijos
the warning DOM my children
‘the warning to my children’
(33) Some no-n-DOM examples
la entrega del / *al paquete
the delivery GEN.DEF / DOM.DEF package
‘the delivery of the package’
la mejora de / *a la constitución
the improvement GEN / DOM the constitution
‘the improvement of the constitution’
el fichaje de / *a Neymar
the hiring GEN / DOM Neymar
‘the hiring of Neymar’
el transporte de / *a los pasajeros
the transportation GEN / DOM the passengers
‘the transportation of passengers’
la crianza de / *a cerdos
the breeding GEN / DOM pigs
‘the breeding of pigs’
I propose the following claim:
(34) The nouns that accept n-DOM on their internal argument are nouns that denote a process event structure, that is, an event structure that does not entail a change of state in the internal argument.
Let’s look at some examples to see how (34) plays out. Consider inspección and colonización. A colonization entails a change of state of the colonized people or place. An inspection does not entail a change of state in that which is inspected. Consequently, inspección accepts n-DOM while colonización does not.
la inspección a las tropas
the inspection DOM the troops
‘the inspection of the troops’
la colonización de / *a los pueblos indígenas
the colonization GEN / DOM the people indigenous
‘the colonization of indigenous peoples’
Similarly for destrucción and ataque: if something has been destroyed, it certainly has undergone a change of state, but being attacked does not entail such a change (and likewise for mutilation vs. torture, capturing vs. chasing, breaking vs. hitting, cleaning vs. sweeping). In every case, the noun whose denotation entails a change of state in the internal argument is the one that does not accept n-DOM.
The generalization in (34) provides insight into what are otherwise puzzling (and subtle) empirical facts. Note the contrast between apuñalamiento and puñalada.
la puñalada a César
the stabbing DOM Caesar
‘the stabbing of Caesar’
el apuñalamiento de / *a César
the stabbing GEN / DOM Caesar
‘the stabbing of Caesar’
Both apuñalamiento and puñalada are translated as ‘stabbing’, but only puñalada accepts n-DOM. This follows from (34): puñalada does not entail that the victim was in fact hurt; the sentence ‘Maria suffered a puñalada but was not hurt because she had a thick vest on’ is not contradictory. On the other hand, apuñalamiento does entail that the victim suffered a stabbing wound.
The contrast in (37) is equally intriguing.
el empujón al carro
the push DOM.DEF cart
‘the pushing of the cart’
*el empujón al carro hacia la tienda
the push DOM.DEF cart toward the store
el empujón del carro hacia la tienda
the push GEN.DEF cart toward the store
‘the pushing of the cart toward the store’
As these examples show, the nominal empujón ‘push’ allows for n-DOM on the object, unless there is a PP that denotes a path or a terminus. Interestingly, this follows directly from (34). Pushing does not entail a change of state and consequently empujón allows for n-DOM. However, when the nominalization includes a path, the resulting structure does entail a movement of the theme object, which qualifies as a form of change of state. Consequently, n-DOM becomes impossible.
(34) can help us understand why ditransitives and unaccusatives do not accept n-DOM. Ditransitive predicates are constructed around the “transfer” schema: the theme is transferred either toward a goal or into someone’s possession. Thus, the theme of ditransitives always entails some change of state in the form of a change of possession or a change of location (see Croft 1998, Levin and Rappaport Hovav 2005). Sometimes this transfer is somewhat abstract, as in “the teaching of math to John” where “math” does not accept n-DOM. But I think that allowing for the possibility of abstract transfer does not threaten generalization (34). Abstract transfer might also account for odd cases such as memorización ‘memorization’ and traducción ‘translation’, which do not accept n-DOM.
As for unaccusatives, the literature agrees that they always entail a change of state (see Levin and Rappaport Hovav 1995, 2005 for overviews): either as creation (‘emerge’, ‘bud’, ‘sprout’), as transformation (‘redden’, ‘widen’), or as displacement (‘rise’). It should follow that they do not accept n-DOM.4
(34) makes a distinction between events that “entail a change of state in the internal argument” and events that do not. This distinction has never been discussed in the context of case assignment, as far as I know. According to the summary presented by Richardson (2012), telicity figures prominently in studies on case variation of internal arguments. Usually, telicity is connected to accusative case (well-known examples are Finnish and Hungarian). However, telicity is too broad a notion for the analysis of the n-DOM data, and I am not aware of any discussion of case variation based on change of state rather than telicity. Let’s consider, as an alternative to hypothesis (34), that the defining property of n-DOM nominals is atelicity. Notice that a predicate like “inspect” should be construed as telic: there is a process and a culmination state, when the inspection is finished. The usual telicity tests corroborate this impression; for instance, the verb “inspect” and the noun “inspection” are compatible with adjuncts that denote a time boundary, as shown by ‘The sergeant inspected the troops in 15 minutes’, ‘the inspection of the troops in 15 minutes’. According to the atelicity hypothesis, the Spanish noun inspección ‘inspection’ should not allow for n-DOM—but it does.
la inspección a las tropas en 15 minutos
‘the inspection DOM the troops in 15 minutes
‘the inspection of the troops in 15 minutes’
On the other hand, ‘inspect’ does not entail a change of state in the internal argument. Thus, the change-of-state hypothesis in (34) correctly predicts the possibility of n-DOM with ‘inspection’.
Within n-DOM nominals, however, there seems to be an additional division. Many of them require n-DOM on the internal argument: for example, ataque ‘attack’, empujón ‘push’, puñalada ‘stabbing’. Some, on the other hand, seem to also accept de optionally. Thus, inspección de las tropas ‘inspection of the troops’ accepts a reading in which de las tropas is the internal argument. I put aside this distinction between optional and obligatory n-DOM for the time being and take it up again in sections 4 and 5.
We may use Levin’s (1999) insights for a deeper understanding of n-DOM. Levin provides a detailed discussion of change-of-state event structures (complex event structures, in her terminology) and process event structures (a subset of her simple event structures). She argues that the former event structure requires an internal argument: all change-of-state predicates (e.g., ‘break’, ‘open’, ‘kill’) have an internal argument because it is required by the event structure. By contrast, the process event structure does not require an internal argument. Some process predicates require an internal argument (‘sweep’ and ‘hit’ do, while ‘run’ does not); this internal argument is required by the root (the constant, in Levin’s terminology), not by the event structure itself.
Levin argues that another distinguishing property follows from this. Whereas the internal argument of change-of-state predicates is always a regular direct object with structural accusative case, the internal argument of process predicates surfaces crosslinguistically in a range of forms: sometimes it is a regular direct object, but sometimes it appears as a prepositional phrase or as a noun phrase bearing some form of inherent or lexical case (dative, instrumental, etc.). It seems clear that the n-DOM phenomenon provides an empirical extension to Levin’s generalization. I take it that n-DOM is one of those special forms of licensing arguments of process predicates. Levin’s observations are based only on data from verb phrases; it is of some interest to find out that they are useful in the analysis of noun phrases as well.
3.2 Syntactic Analysis
In this section, I provide an account for the generalization in (34). The intuition that I formalize is that the complexity of the event affects the availability of n-DOM. Change-of-state events are more complex than process events. Change-of-state events include both a process and a resultant state while process events only include a process—and this is the crucial factor.
Following recent developments (see in particular Alexiadou 2001, Ramchand 2008, Borer 2012, 2013a,b), I take it that event structure constitutes a syntactic structure. If so, the complexity of the event structure is reflected in the syntax, which should correspondingly be more complex for more complex events. Change-of-state events have a complex syntax with separate heads for the two subevents, the process and the result, while process events only include a head for the process. In a nutshell, I argue that the extra layer of structure in change-of-state events protects the internal argument from being assigned n-DOM. A second rule of genitive case assignment may then apply to internal arguments in change-of-state events.
The literature offers a number of proposals regarding event syntax. To approach the empirical problems of concern here, I adopt assumptions from various sources, in particular Ramchand 2008. I do not attempt to provide a theory that derives the different verb classes because that would take us too far afield. In particular, given my focus on the n-DOM phenomenon, I consider only transitive predicates.
Let’s start with the following skeleton:
(39) Init [ProcP IA Process [√]]
The head Initiation (Init; borrowed from Ramchand 2008) initiates the event and introduces the external argument (EA), which might be an agent, a causer, or just an origin; Init comes in different “flavors.” The head Process denotes the alteration in the environment triggered by the external argument, the actual event, which has the potential of impinging on the internal argument (IA). In this structure, Process is a two-place predicate whose arguments are the root and the internal argument (again, I am only considering transitive predicates). Examples are sweeping the floor, kissing his cheek, hitting the ball, praising the athlete, and so on.
Now consider the following structure:
(40) Init [ProcP Process [sP IA [s′ State √]]]
In (40), Process selects for a resultant state. The head s denotes this resultant state and selects a root as a complement. It merges with the internal argument, which becomes the subject of this new state.
I need to articulate an additional assumption. First, event structure is not necessarily introduced by a verbal head, as is commonly assumed; rather, it is a cross-categorial property (see López 2015). Since event structure is independent of category, (39) and (40) could be either a noun or a verb. I suggest that a category label is merged in the structure and takes Process as a complement. Once the structure is defined as nominal, it can be selected by the regular nominal heads, in particular D. Finally, I assume that a complete noun phrase also includes a K head that may spell out as a case morpheme. Thus, an event nominal that involves a change of state looks like this:
(41) K [D [Init [n [Process [State √]]]]]
Notice that the structure Init[n] has obvious parallels in the contemporary literature. Alexiadou, Anagnostopoulou, and Schäfer (2006), Harley (2013), and Legate (2014) separate the verbalizer v from the head Voice and let the latter select for the former in the structure Voice[v]. The proposed structure Init[n] largely incorporates these ideas.
The following summarizes the structure fragments of interest here:
[Init (KEA ) Init [n [p KIA p [√]]]]
p = process
√ = root (i.e., √ATAQ-, √TORTUR-, √PERSECU-, . . .)
[Init (KEA ) Init [n [p p [s KIA s [√]]]]]
s = state
√ = root (i.e., √CONSTR-, √MUTIL-, √ROMP-, . . .)
Event structures are compressed into something smaller before spell-out. I assume that head movement ensures that the root, the event heads, and the categorizing head stand as a word unit for the purposes of Vocabulary Insertion. A complete structure of an event nominal with n-to-Init movement is shown in (43).
Notice that in these structures I have enclosed the external argument in parentheses. The question is whether there is an external argument in nominalizations at all. In particular, Picallo (1991) argues that nominalizations in Catalan are passive, while Alexiadou (2001) argues quite extensively that nominalizations are crosslinguistically always ergative. One of the arguments presented is that an external argument cannot appear as an of-genitive; instead, it always appears as a by-adjunct or as an s-gen in English. Consider the examples in (44).
the destruction of the city by Napoleon
*the destruction of Napoleon (of the city)
Napoleon’s destruction of the city
the deliberate destruction of the city
(44) exemplifies a change-of-state nominal. In (44a), the external argument appears in an adjunct phrase. (44b) shows that indeed the external argument is not allowed in the structure as an argument noun phrase in genitive case. In (44c), the external argument appears in the s-gen construction. However, Alexiadou argues, the s-gen element can be an argument or an adjunct; she suggests that we interpret Napoleon as the agent of destruction in (44c) because of our encyclopedic knowledge and not because a θ-role has been assigned to it. (44d) shows that some form of implicit argument that licenses a subject-oriented adjunct must be present in these nominalizations. Examples similar to (44a–b,d) with identical judgments can be constructed in Spanish. I believe Alexiadou’s argument holds true, but only of change-of-state nominals.
Now consider the examples in (45). Here we have a process nominal, and the internal argument exhibits n-DOM. Notice that an external argument in genitive case is possible here, as shown in (45b–c). In particular, notice that (45c) (repeated from (3c)) shows that the genitive argument can bind a reflexive in the internal argument.
el deliberado ataque a la ciudad
the deliberate attack DOM the city
‘the deliberate attack on the city’
el ataque de Juan a la ciudad
the attack GEN Juan DOM the city
‘Juan’s attack on the city’
el ataque de Juan a sí mismo
the attack GEN Juan DOM himself
‘Juan’s attack on himself’
Thus, I take it that there is no general ban on an external argument in nominalizations. Rather, some property of change-of-state nominalizations prevents an external argument with genitive case. I will return to this issue in section 4.1.
Given the structures in (42), I propose the rules in (46) to account for the distribution of structural case in Spanish argument-taking nominals. Further, I adopt the Case Filter as an active grammatical principle; it ensures that any KP must be in a dependency relation with a case assigner.
Following Rule 1 in (46a), n-DOM is assigned to an internal argument by n in a government configuration. I adopt a basic notion of government: a syntactic terminal α. governs a constituent β if α. c-commands β and there is no other syntactic terminal γ such that γ c-commands β and α. c-commands γ.
(47) α [β → α governs β
(48) α [γ [β → α does not govern β
Thus, n governs Spec,Process because no other syntactic terminal stands between them. Rule 2 in (46b) is the spell-out rule for genitive case. Genitive case shows up on many DP constituents (modifiers, possessors, relational nouns); I assume there is an extra rule (or several rules) for assigning genitive case to these constituents. I do not discuss this further here.
Returning to (46), the crux of the matter is this: the n-DOM rule only applies to an argument that is governed by n. It follows that an internal argument in Spec,s is too far away to feed it. This explains the absence of n-DOM on internal arguments that suffer a change of state7 and accounts for the first puzzle, the contrast between ataque and captura.8
Thus, the intuition that the rules in (46) try to formalize is the following: within the Spanish nominal there is a general rule of genitive case assignment, with its locus in D, which reaches any argument within the nominal. Versions of this rule are widespread crosslinguistically. Additionally, there is a more specific rule of case assignment that resides in n and applies more locally, to an argument that is structurally adjacent to it. I find the existence of these two rules of case assignment with different scopes to be necessary to account for n-DOM in Spanish as well as to resolve the two other puzzles mentioned in section 1; this will become clear as we proceed.
Before I continue, let me show why I have not chosen other, more mainstream paths for the analysis. First, why have I used government as a locality condition? I am aware that the notion of government was largely abandoned about twenty years ago (although it often returns through the back door) and that nowadays, locality requirements are formulated in terms of phases. Could (46a) be rephrased in terms of phases? One could posit that sP is a phase and that that is the reason why n cannot reach it. However, this would leave D out of the sP phase too, and therefore genitive case on the argument in Spec,State would not be possible.
Second, might the two case morphologies be due to different flavors of n, in parallel to Harley’s (1995) and Arad’s (1999) influential approach to vP structure? In this view, we would have an nCoS that does not assign n-DOM and an nProc that does. This solution would work technically and would be reasonably simple. However, we would lose the intuition that the nominals that do not allow n-DOM have a more complex event structure than those that do. The appearance of n-DOM on one class of nominals rather than the other would be mere chance and not related to any structural reason.
3.3 Psych Predicates
The previous discussion could lead to the conclusion that n-DOM is a form of inherent case, associated exclusively with the semantic role of undergoer, the subject of Process. However, n- DOM can also show up in stative psych predicates, where there is no process. This leads to the conclusion that n-DOM is a structural case.
A number of psych predicates also allow or require n-DOM. The following are three examples:
el amor de Juan a / por María
the love GEN Juan DOM / by Maria
‘Juan’s love for Maria’
(cf. Juan ama a María.
Juan loves DOM Maria)
el miedo de Juan a / ??por las tormentas
the fear GEN Juan DOM / by the storms
‘Juan’s fear of storms’
la admiración de Juan a / por los catalanes
the admiration GEN Juan DOM / by the Catalans
‘Juan’s admiration for Catalans’
(cf. Juan admira a los catalanes.
Juan admires DOM the Catalans)
The examples in (49) involve psych nouns of a very canonical transitive type: the experiencer is the subject of the DP and the theme/stimulus is the internal argument. Other nouns with the same structure are (among others) respeto ‘respect’, odio ‘hatred’, resentimiento ‘resentment’, temor ‘fear’, aborrecimiento ‘loathing’. Notice that the theme may also appear in a PP headed by por ‘by’.
The analysis formalized above can be extended to these examples. I take it that a psychological state has the structure in (50). The construction includes a State but no Process, and the external argument of the initiator is the experiencer (for extensive discussion of psych predicates, see Landau 2010 and references therein; for discussion of psych nouns, see Fábregas, Marín, and McNally 2012).
Stative psych nouns
(EA) Init [n [sP IA s √]]
√ = root (i.e., √AM-, √MIED-, √ADMIR-, . . .)
In representation (50), the internal argument is governed by n and can receive n-DOM. However, there are other psych nouns that do not accept n-DOM.
la vergüenza de Juan por / *a María
the shame GEN Juan by / DOM Maria
‘Juan’s shame about Maria’
(cf. A Juan le avergüenza María.
DAT Juan CL shames Maria
‘Juan is ashamed of Maria.’)
el asombro de Juan por / *a María
the amazement GEN Juan by / DOM Maria
‘Juan’s amazement over Maria’
(cf. A Juan le asombra María.
DAT Juan CL amazes Maria
‘Maria amazes Juan.’)
el interés de María por / *a las matemáticas
the interest GEN Maria by / DOM the mathematics
‘Maria’s interest in mathematics’
(cf. A María le interesan las matemáticas.
DAT Maria CL interest the mathematics
‘Maria is interested in mathematics.’)
Other nouns of this type are cabreo ‘anger’, crispación ‘tension’, enfado ‘anger’, enfurecimiento ‘fury’, excitación ‘arousal’, susto ‘fright’, sorpresa ‘surprise’. Some of these psych predicates could be regarded as causative: for instance, susto ‘fright’ and sorpresa ‘surprise’ (Pesetsky (1995) and Arad (1999), among others, have proposed a causative analysis for this type of psych predicate). Under this causative assumption, the structure of these nominals would include a process as well as a resultant state and therefore the analysis presented above for change-of-state predicates (see (42)) generally would apply. However, a causative analysis is less plausible for some of these nouns, such as vergüenza ‘shame’ and interés ‘interest’.
Is there a generalization that would allow us to predict which psych nouns accept n-DOM and which do not? In fact, there is one. The verbal counterparts of the psych nouns that do not accept n-DOM are unaccusative (see Belletti and Rizzi 1988 for the original description of unaccusative psych verbs): the experiencer appears in dative case and the theme/stimulus in nominative case. As for the psych nouns that accept n-DOM, some have verbal counterparts that accept v-DOM (e.g., amor ‘love’, admiración ‘admiration’), while others do not have a verbal counterpart at all (e.g., miedo ‘fear’)—and examples like the latter lead us to discard the possibility of identifying n-DOM with v-DOM (see also section 2.3).
At this point, all I can say is that it seems that there are some psych roots like √INTERÉS- and √ASOMBR- that bear the property of canceling case assignment by n or v. As a consequence, the nouns derived from these roots do not allow n-DOM, while the verbs derived from them do not allow accusative case. This seems to be an instance of a more general property, since governed prepositions work in the same way: from the root √CONFI- ‘trust’ we can derive the verb confiar and the noun confianza, and both select the preposition en. A deeper exploration of this property and its consequences for the theory of grammar must await future research.9
Here is the interesting conclusion that we can extract from the psych data: n-DOM is not a form of inherent case assigned together with a θ-role. n-DOM is assigned to any constituent that is governed by n in a nominalization; n-DOM is a form of structural case.
4 The Second and Third Puzzles
4.1 The Second Puzzle
The second puzzle turns on the distribution of genitive case. Rule 2 in (46) would lead us to expect that the external argument should be able to appear freely in genitive case. However, this is not so: the availability of genitive case on the external argument reflects an additional empirical difference between change-of-state and process nominals. In change-of-state nominals, the genitive argument is always the internal argument. For example:
(52) Change of state
el escalamiento del conflicto
the escalation GEN.DEF conflict
‘the escalation of the conflict’
la evaporación del gas
the evaporation GEN.DEF gas
‘the evaporation of the gas’
la colonización de los indígenas
the colonization GEN the indigenous.peoples
‘the colonization of indigenous peoples’
la humillación del alumno
the humiliation GEN.DEF student
‘the humiliation of the student’
el apuñalamiento de César
the stabbing GEN Caesar
‘Caesar’s stabbing (by someone)’
If there is a constituent that looks like an external argument in change-of-state nominals, it is an adjunct introduced by por/by. Thus, change-of-state nominals must have some property that prevents application of Rule 2 to the external argument.
In process nominals, the state of affairs is more complicated. As I mentioned earlier, some process nominals require a to mark the internal argument while others allow a or de. This gives rise to an ambiguous reading for the genitive argument among the latter nominals. Consider the examples in (53). Persecución ‘pursuit’ and despedida ‘farewell’ allow de or a on the internal argument. Consequently, (53a) and (53b) are ambiguous, since the de-argument could be an external or an internal argument. As for (53c–e), estafa ‘swindle’, advertencia ‘warning’, and puñalada ‘stabbing’ only allow n-DOM on the internal argument and, as a result, the de-argument can only be the external argument.
la persecución de los fugitivos
the pursuit GEN the fugitives
‘the pursuit of the fugitives’ (ambiguous)
la despedida del empleado Pérez
the farewell GEN.DEF employee Pérez
‘the farewell of the employee Pérez’ (ambiguous)
la estafa de Rita
the swindle GEN Rita
la advertencia del policía
the warning GEN.DEF police.officer
‘the police officer’s warning’
la puñalada de Bruto
the stabbing GEN Brutus
‘Brutus’s stabbing (of somebody)’
Thus, the second puzzle actually involves three questions:
P2Q1: Why do change-of-state nominals not accept an external argument in genitive case?
P2Q2: Why do process nominals allow the external argument to appear in genitive case?
P2Q3: Why do some process nominals allow an internal argument in genitive case?
Let’s start with P2Q1. Examples (54a–c) show that the only way to introduce an overt external argument with change-of-state nominals is in the shape of a por-adjunct.
la captura de los fugitivos
the capture GEN the fugitives
‘the capture of the fugitives’
*la captura de la policía de los fugitivos
the capture GEN the police GEN the fugitives
la captura de los fugitivos por la policía
the capture GEN the fugitives by the police
‘the capture of the fugitives by the police’
I take (54b) as the starting point of the analysis. As Fábregas (2013c) discusses extensively, the restriction on the presence of two genitives in the same noun phrase as in (54b) can be accounted for using the DC (see (26)). Since the DC prevents a fully formed transitive predicate, the grammar adopts an ergative structure, such that the external argument is in fact an adjunct by-phrase, as shown in (54c). Under the assumption that change-of-state nominals generally adopt an ergative structure, it follows that in (54a) de los fugitivos can only be an internal argument. Thus, the answer to P2Q1 turns around the DC.
As mentioned in section 3, Alexiadou (2001) has argued extensively that all nominalizations are ergative structures. However, this claim is too general. As I show in (45), (53), and below, process nominals do seem to have a proper transitive structure with an external argument. Moreover, Alexiadou presents the ergative structure of nominalizations as a primitive property; however, if my analysis is correct, it is derived from the DC.
Let’s move on to P2Q2: why do process nominals, unlike change-of-state nominals, allow the external argument to appear in genitive case? The question can now be addressed as follows. The external argument can receive genitive case from D, following Rule 2 in (46). The DC, which forces change-of-state nominals to project the external argument as an adjunct, does not have the same effect among process nominals. This is because in process nominals, the internal argument can bear a distinct morphology on K via Rule 1, thus avoiding a violation of the DC.
To see how this works in detail, consider inspección ‘inspection’. This is a nominal whose genitive argument can be external or internal. Consider the examples in (55).
la inspección de los soldados
the inspection GEN the soldiers
‘the inspection of the soldiers’ or ‘the inspection by the soldiers’
*la inspección de la policía de los soldados
the inspection GEN the police GEN the soldiers
la inspección de los soldados por la policía
the inspection GEN the soldiers by the police
‘the inspection of the soldiers by the police’
la inspección a los soldados
the inspection DOM the soldiers
‘the inspection of the soldiers’
la inspección de la policía a los soldados
the inspection GEN the police DOM the soldiers
‘the inspection of the soldiers by the police’
la inspección a los soldados por la policía
the inspection DOM the soldiers by the police
‘the inspection of the soldiers by the police’
Since inspección is a nominal that allows but does not require n-DOM, (55a) is ambiguous: the soldiers may be the ones who inspect or the ones who are inspected. (55b) is ungrammatical, which can again be attributed to the DC. The introduction of a por-argument in (55c) forces an internal argument reading on the genitive KP. (55d–f) are examples with n-DOM. In (55d), the n-DOM is the only argument in the noun phrase; in (55e), the genitive argument is the external argument; and in (55f ), the external argument appears as an adjunct. The data in (55) seem to form a complex patchwork, but we can immediately extract the following generalizations: (a) the genitive argument can be external or internal; and (b) this ambiguity can be resolved in context, so that if there is an n-DOM argument, the genitive must be an external argument, and if there is a by-adjunct, the genitive must be an internal argument.
We can now tackle P2Q2: why process nominals allow a genitive external argument. The answer lies in the availability of n-DOM, which allows process nominals to project a full transitive structure without running afoul of the DC. The key datum is (55e): here, it is possible to have a genitive argument as a proper external argument provided that the internal argument is introduced by n-DOM.
For completeness, let’s consider the data for ataque ‘attack’, a nominal that requires n-DOM on its internal argument.
el ataque de los fugitivos
the attack GEN the fugitives
‘the attack by the fugitives’
*el ataque de la policía de los fugitivos
the attack GEN the police GEN the fugitives
*el ataque de los fugitivos por la policía
the attack GEN the fugitives by the police
el ataque a los fugitivos
the attack DOM the fugitives
‘the attack on the fugitives’
el ataque de la policía a los fugitivos
the attack GEN the police DOM the fugitives
‘the attack on the fugitives by the police’
el ataque a los fugitivos por la policía
the attack DOM the fugitives by the police
‘the attack on the fugitives by the police’
(56a) is not ambiguous: de los fugitivos must be an external argument. It follows that (56c), in contrast to (55c), should also be ungrammatical. The other sentences work as in the inspección example. The obligatoriness of the external argument reading in (56a) is discussed in section 5.
Let’s now tackle P2Q3. P2Q3 asks why the genitive argument of some process nominals can be interpreted as an internal argument—that is, why (55a) is ambiguous. There are two possible paths to take here. According to the first path, a is always the spell-out of Rule 1 and de is always the spell-out of Rule 2. Thus, there would be nothing surprising about the ambiguity of (55a); it would just mean that internal arguments may be subject to Rule 2. But this path is problematic for two reasons. First, we would have to figure out what inhibits application of Rule 1 so that application of Rule 2 on an internal argument is permissible. Second, we would have to figure out why Rule 2 does not apply in the subset of process nominals that do not allow internal arguments to surface in genitive case (exemplified with ataque ‘attack’ in (56)). These two problems are insurmountable and therefore I do not take this path.
According to the second path, the internal arguments of process nominals are always subject to Rule 1. The reason why some nominals optionally surface with a de-form is that de is a possible spell-out of Rule 1 for this subset of nominals. Since de is a possible spell-out for Rule 1 for a subset of nominals, it follows that de los soldados in (55a) may be interpreted as an internal argument. I adopt this path here—and it will prove to have additional advantages in section 4.2. Let’s see how this works out. Assume that application of Rule 1 excludes application of Rule 2. This can be achieved if we take the rules to be ordered so that Rule 1 applies before Rule 2. This is natural in a bottom-up theory of grammar in which n is introduced into the structure before D is, as in (57).
n [KIA p √]
Rule 1: n assigns case to K.
D [Init n [KIA p √]]
Rule 2: D does not assign case to K because the latter is already case-marked.
To summarize this section: The account of the difference between change-of-state nominals and process nominals capitalizes on my earlier conclusion that process nominals have two strategies for structural case assignment (Rules 1 and 2) while change-of-state nominals only have one (Rule 2). The reason why change-of-state nominals cannot make use of Rule 1 is that the event involves a more complex syntactic structure. Since process nominals have two sources of structural case, they can have a proper external argument without violating the DC, as shown in (55) and (56).
4.2 The Third Puzzle
Let’s now turn to the third puzzle: Anderson’s (1977) realization that the internal argument must be affected when used in the Saxon genitive or, more generally, in the s-gen. Consider the examples in (58).
John’s / his capture / improvement / transformation / hiring of Bill
John’s / his farewell to / criticism of / warning of / inspection of Bill
John’s / his capture / improvement / transformation / hiring
John’s / his farewell / criticism / warning / inspection
As (58a–b) show, when we have both an s-gen and an of-genitive, the former is the agent and the latter is the theme, regardless of the type of nominalization. When there is only one argument and it appears in the s-gen form, the change-of-state predicate requires a theme interpretation (58c), while the process predicate requires an agent interpretation (58d). This particular distribution is what needs an account.
As mentioned in section 1, Spanish does not have a Saxon genitive construction. However, it does have a possessor construction, in which one of the arguments of the nominalization appears as a possessive adjective. When that happens, the pattern of interpretation is exactly as in English. In (59a–b), the possessor is the agent; in (59c), it is a theme; and in (59d), it is an agent. I take it that this pattern does not arise by chance; rather, it reveals that there is a commonality in structure between the nominalizations in the two languages.
su captura de la ciudad
POSS.3 capture GEN the city
‘her / his capture of the city’
su crítica de / a Rajoy
POSS.3 criticism GEN / DOM Rajoy
‘her / his criticism of Rajoy’
‘her / his capture’
‘her / his criticism’
In the previous sections, I argued that the distinction between process nominals and change-of-state nominals underlies an account of the case distribution in Spanish nominals. In this section, I argue that the distribution of s-gen also depends on the complexity of event structure.
Recall that Anderson (1977) argues that the notion relevant to accounting for the restriction on the English s-gen is affectedness: themes can be used in the s-gen construction only if they are affected. Smirnova (2015:571) defines internal arguments in s-gen constructions as “denot[ing] a change of state or a change of location.” I entirely agree with Smirnova’s description. Thus, I would like to foreground the following descriptive generalization: the set of nominals that do not accept n-DOM in Spanish is the same set that forces an internal argument reading on the s-gen. It seems that a parallel analysis is called for.
In order to approach this problem, I draw again on insights provided by Levin (1999). Recall that she argues that change-of-state events require an internal argument, while the internal argument in process predicates is required by the root (the constant, in her terminology). Additionally, she proposes the following condition:
Structure Participant Condition (SPC)
There must be an argument XP in the syntax for each structure participant.
We are now very close to an answer to the third puzzle. As a consequence of the SPC, the event structure of change-of-state nominals requires an internal argument (in the present model: an argument in Spec,State), while the event structure of process nominals does not: if a process nominal has an internal argument (an argument in Spec,Process), it is selected by the root. In this light, consider John’s capture again; if John were an external argument, the nominalization would violate the SPC because the internal argument, required by the event structure, would be absent. So, John can only be the internal argument. As for John’s criticism, the process event structure does not require an internal argument and therefore John can be interpreted as the external argument. Minimality then forces the external argument interpretation. In the following paragraphs, I flesh out the details.
The analysis of s-gen requires that we assume a third rule of case assignment in argument-taking nominals, which we can call Rule 3.
(61) Rule 3: D[poss] assigns [s-gen] to K if Merge (K, D).
Rule 3 is active in Spanish and English. It applies liberally to all kinds of KPs in English, allowing branching KPs to form a Spec,D[poss] while possessive adjectives fuse with the D[poss] . Spanish only has the second possibility: D cannot host a specifier in this language.
Let’s now focus on the Spanish data. I suggest the following approach. The internal arguments of process nominals are always subject to Rule 1, as claimed in section 3.3. The reason why some nominals optionally take a de-form is that de is a possible spell-out for Rule 1. Let’s further assume that Rule 1 applies before Rules 2 and 3. This ordering is a direct consequence of the claim that Rule 1 involves n as the assigner while Rules 2 and 3 involve D (see (57)). As a result, the pronoun su in (59d) cannot be interpreted as an internal argument because the internal argument of crítica ‘criticism’ is subject to Rule 1, which excludes application of Rule 2 or Rule 3.
Exporting this idea to English requires just one assumption: English also has Rule 1 but it does not have a specific spell-out for n-DOM. Therefore, the output of Rule 1 falls within the general spell-out rule for K within a nominal; namely, it is of. This assumption is all that is required to account for (58d): John cannot be the patient of criticism because John is subject to Rule 1, which blocks Rule 3.
Turning now to change-of-state nominals and the data in (59a,c): The internal argument of change-of-state predicates can appear as an of-genitive or as s-gen—that is, it can receive case via Rule 2 or Rule 3. This follows from my assumptions: if both Rule 2 and Rule 3 apply when D is inserted into the structure, then neither blocks the other. This is how the grammatical English phrases the city’s destruction by the soldiers and the soldiers’ destruction of the city come about. On the other hand, *the city’s destruction of the soldiers (with the soldiers as agent) is banned because change-of-state nominals are ergative and therefore of the soldiers is not a possible realization of the external argument.
To conclude: The detailed examination of case patterns in nominalizations has revealed a commonality of structure between English and Spanish that was not apparent at first. The only difference between these two languages turns out to revolve around the spell-out of the output of Rule 1.
One might suggest an alternative analysis of the s-gen phenomena in which English has no Rule 1. However, this analysis is not successful. That is, suppose that English has a Rule 2 that assigns of-genitive, a Rule 3 that assigns s-gen, and no Rule 1. The main challenge of this reduction is how to prevent an internal argument of a process nominal from receiving s-gen (as in (58b,d) and (59b,d)). One option might be that assignment of Rule 2 to the internal argument blocks Rule 3. Consequently, the possibility of the criticism of his policy would block *his policy’s criticism. However, this analysis overgenerates, because the destruction of the city would incorrectly block the city’s destruction. The fact is that of does not block s-gen generally (it only does so with process nominals), and this fact is not captured with a simple system including only Rules 2 and 3. In contrast, my analysis makes the right predictions. Both the city’s destruction and the destruction of the city are ruled in: destruction is a change-of-state nominal and therefore only Rules 2 and 3 are operative. Both rules involve D as head and therefore can apply freely. Criticism instead is a process nominal, and therefore Rule 1 enters the fray: *his policy’s criticism is ungrammatical because Rule 1 has applied and assigned case to the internal argument; this case is spelled out as of, resulting in the criticism of his policy.
Another alternative to the analysis presented here could involve inspecting the argument structure in more detail. For instance, one could argue that process nominalizations are always transitive with an external argument in Spec,Init that could optionally spell out as an empty pronominal pro. Thus, *his policy’s criticism would be blocked by Minimality because pro would intervene. However, the assumption that process nominalizations are always transitive would wrongly predict that the criticism of his policies by the press should be ungrammatical. All in all, it seems to me that assuming Rule 1 in English is the only way to account for the data; it also has the advantage of providing an elegant analysis of the differences and similarities between English and Spanish nominalizations, an analysis based not on disparities over how case morphology is computed but only on how case morphology is spelled out.
To summarize: Section 4.1 took up the second puzzle, namely, why the external argument can appear in genitive (de/of ) case in process nominals and not in change-of-state nominals. This property is derived from the simpler event structure of process nominals. Process nominals can give rise to n-DOM, which itself allows for a genitive external argument without violating the DC. Section 4.2 then addressed the third puzzle, regarding s-gen nominals and the affectedness constraint; this constraint was also shown to follow from event structure, as the simple event structure of process nominals makes the internal argument susceptible to Rule 1 and therefore inaccessible to Rule 3, the rule that assigns s-gen. Thus, event structure has wide-ranging consequences for the distribution of morphological case in nominalizations.
5 Obligatory and Optional n-DOM
In the previous sections, I have noted in passing that some process nominals exhibit n-DOM obligatorily while others do so optionally. In this section, I address this issue and present a tentative description of the optional and obligatory n-DOM nominals.
el ataque de Napoleón a la ciudad
the attack GEN Napoleon DOM the city
‘Napoleon’s attack on the city’
*el ataque de la ciudad por Napoleón
the attack GEN the city by Napoleon
el ataque a la ciudad por Napoleón
the attack DOM the city by Napoleon
‘the attack on the city by Napoleon’
la inspección de los ingenieros a las cañerías
the inspection GEN the engineers DOM the pipes
‘the engineers’ inspection of the pipes’
la inspección de las cañerías por los ingenieros
the inspection GEN the pipes by the engineers
‘the inspection of the pipes by the engineers’
la inspección a las cañerías por los ingenieros
the inspection DOM the pipes by the engineers
‘the inspection of the pipes by the engineers’
The conclusions reached in section 4 allow us to be more precise with respect to the difference between (62b) and (63b): ataque requires that the output of n-DOM always spell out as a, while inspección has two alternative spell-outs for Rule 1.
In fact, the division of process nominals into two classes is insufficient. I claim that there are two types of obligatory n-DOM nominals. Consider the following lists:
advertencia ‘warning’, consejo ‘advice’, acoso ‘harassment’, alabanza ‘praise’, obediencia ‘obedience’, acatamiento ‘compliance’
ataque ‘attack’, bofetada ‘slap’, patada ‘kick’, golpe ‘blow’, barrida ‘sweep’, puñetazo ‘punch’, beso ‘kiss’, abrazo ‘hug’, empujón ‘push’, achuchón ‘cuddle’, combate ‘combat’
inspección ‘inspection’, repaso ‘revision’, persecución ‘persecution, pursuit’, despedida ‘farewell’, reconocimiento ‘recognition, acknowledgment’
The obligatory n-DOM(1) list consists of predicates whose internal argument is necessarily human. Since v-DOM requires that the complement be animate, there seems to be a connection between n-DOM and v-DOM in this respect (see section 2.1). The connection is weak: if the nominal predicate does not require a human internal argument, it does not fall in the n-DOM(1) camp, while v-DOM requires that the DP be animate even if the verbal predicate does not require an animate complement. The presence of obediencia ‘obedience’ and acatamiento ‘compliance’ might seem surprising here, since one can oebey or comply with the law, not necessarily a human being. However, these nouns confirm the generalization: it is possible to say obediencia a la ley or obediencia de la ley ‘obedience to the law’ but it is not possible to say *obediencia de los padres ‘obedience to parents’, unless the genitive argument is an external argument. So with this subset of nominals, referring to a as a form of DOM is appropriate.
The contrast between n-DOM(2) and n-DOM(3) is puzzling. To see this, first consider Beavers’s (2011) affectedness hierarchy (AH), shown in (67). (For present purposes, I have reduced Beavers’s AH from four members to three.)
Internal argument undergoes a change:
A quantized change: break, shatter, destroy, . . .
A nonquantized change: widen, cool, cut, . . .
Internal argument has potential for change (because there is surface contact, impact):
wipe, scrub, rub, punch, hit, . . .
Internal argument is unspecified for change:
see, smell, follow, . . .
The internal arguments noted in (67.1) are the highest ones on the AH. They are also the ones that do not accept n-DOM—that is, the ones that are not subject to Rule 1. The second arguments on the AH (67.2) are the ones that take an obligatory n-DOM(2). Finally, the ones lowest on the AH (67.3) take optional n-DOM(3). If we look at the distribution of n-DOM from a scales perspective (as in Aissen’s (2003) analysis of v-DOM), the result is puzzling. If the AH is a scale, we would expect its intermediate member to behave in a manner intermediate between the other two. Instead, the intermediate member is the most radical one.
Let’s try—somewhat tentatively—a second approach. I think the correct generalization is this: the predicates involved in n-DOM(2) are all of the type in which the internal argument is subject to surface contact or impact by the external argument (although the contact might be somewhat abstract in the Internet era, as in the case of ataque). n-DOM(3) does not entail any physical contact.
As just noted, n-DOM(2) nominals involve an internal argument that denotes a surface that is contacted or impacted. Interestingly, the directional locative preposition in Spanish also spells out as a.
(68) ir a ‘go to’, venir a ‘come’, dirigirse a ‘go in the direction of ’, llegar a ‘arrive at’, . . .
Thus, I suggest that the reason why n-DOM(2) is obligatory is that the use of a in this context is reinforced by the similarity of both the n-DOM morpheme and the θ-role of the internal argument to the preposition and the θ-role of the locative construction.
To conclude: Argument-taking nominals with a process argument structure are compatible with n-DOM. If the nominal selects for an animate complement, Rule 1 spells out as a obligatorily. If the nominal entails contact with a surface, Rule 1 also spells out as a. Otherwise, the spellout of Rule 1 as a is optional.
In sections 1 and 2, I argued that Spanish n-DOM is distinct and independent from v-DOM. In particular, n-DOM is not a survivor of structural or inherent accusative case brought from the vP into the nominalization. It is however clearly a form of structural case, since it can be assigned to arguments that bear different θ-roles. Consequently, n-DOM is a previously unidentified grammatical phenomenon.
In sections 3 and 4, I discussed in detail three puzzles concerning the grammar of case in argument-taking nominals: the distribution of n-DOM in Spanish, the interpretation of genitive arguments in Spanish and English, and the affectedness constraint on internal arguments in the s-gen construction in English. I argued that all three puzzles can be accounted for by means of an analysis that capitalizes on the syntactic structure of events. In particular, I argued that the more complex event structure of change-of-state nominalizations gives rise to important differences in the case distribution of their arguments. My analyses exploited proposals for the syntax of events that have previously only been argued for in the context of verb phrases. I assumed the essential similarity of what Ramchand (2008) calls first-phase syntax in verb phrases and event-denoting noun phrases. Likewise, I assume that the grammars of Spanish and English are distinct only with respect to one spell-out rule: both the structure and the features involved are the same.
My analyses naturally lead to some questions regarding the theory of morphological case. When case theory became central in generative linguistics in the early 1980s, it was commonly assumed that the morphological case found in noun phrases was genitive case, which was regarded as a type of inherent (not structural) case (see Chomsky 1981), even though genitive case can be associated with any θ-role. Certainly, linguists were aware that a variety of morphological cases can be found in a noun phrase, but the fact of the matter is that the study of cases within the noun phrase has been postponed for a long time. The outstanding exception to the latter claim is Pesetsky’s (2013) detailed study of case in the Russian noun phrase.
It seems to me that one of this article’s contributions is precisely its confirmation that structural case is assigned within the noun phrase. But this realization, together with the particular analyses developed here, presents new puzzles for our understanding of morphological case. This article has shown that the way case assignment operates in a nominalization seems to be bewilderingly different from the way it operates in a clause. In a clause (putting aside ergative languages), there are two clearly structural cases—nominative and accusative—each of which is assigned within its phasal domain. In particular, nominative is assigned by T, accusative by Voice/Init. Moreover, in a clause, accusative is a dependent case assigned by the same head that assigns the external θ-role. In addition, a clause can include inherent cases of various types. In a nominalization, there is a general operation of genitive case assignment that affects external or internal arguments and is limited only by event structure. Voice/Init assigns no case; instead, a rule of case assignment that resides in n assigns case; case assignment by n takes place within its government domain, rather than within the broader domain of a phase. I hope that future research will explain why case assignment in nominalizations appears to be so different from case assignment in clauses.
1 The example ataque ‘attack’ and many of the Spanish nouns used in this article are probably not derived from verbs; rather, both the noun and the verb are derived from the same root. I will refer to them occasionally as “nominalizations” even if properly speaking they are not. Fábregas (2013a) and I (López 2015) have shown that Spanish nominals directly derived from the root can be argument-taking complex event nominals.
2 This is true for most varieties of Spanish, which follow the so-called etymological distribution for pronominal case. As is well-known, some peninsular varieties follow a distribution of direct/indirect object pronouns based on animacy and gender rather than grammatical function.
3 Notice that the dative case marker is homophonous with DOM, a matter that has inspired some analyses of DOM as a form of dative. In this article, I adopt the relatively conservative idea that they are two distinct morphemes. Likewise, the directional preposition is also spelled out as [a].
4 Nevertheless, notice that (34) is not stated as a double conditional. That is because there are some process nominals that do not accept n-DOM. Some of them involve governed prepositions: for instance, confianza ‘trust’ governs the preposition en (as does the verbal derivation confiar ‘trust’). A few others remain unaccounted for: analisis ‘analysis’, evaluación ‘evaluation’, abandono ‘abandonment’, deserción ‘desertion’. For instance, my consultants rejected la evaluación a los maestros ‘teachers’ evaluation’ (although I found some examples in a Google search).
5 The scholars who propose a Voice-v separation argue that the head that introduces the external argument and assigns case to the object is Voice. This assumption allows for an elegant account of the active/passive alternation: if Voice is absent or in a reduced form, it follows that there will be no external argument or accusative case in the structure. If my analysis of Init-n were truly parallel, Init would be a case assigner, not n. Since I argue that n and not Init is a case assigner, I am led to conclude that there is a difference between the nominal and verbal domains regarding the properties of case assignment. The empirical evidence supports this conclusion: we have already seen that assignment of n-DOM is compatible with a passive-like structure (see (13)), which suggests that n-DOM assignment cannot depend on the transitivity of Init/Voice.
6Pesetsky (2013) suggests that genitive case originates in the noun instead. I have no strong arguments for the view that genitive case is dependent on D, except that, since genitive case can affect any argument in the clause, it should probably be dependent on a high functional head. The literature on possessors and genitive case assumes or argues that genitive case is assigned from a high position in the DP structure (see Szabolcsi 1994 for an early summary of the arguments).
7 In Ramchand 2008, the subject of State raises to Spec,Process, so that the argument that changes state is also the one that undergoes a process. Under this analysis, the structural distinction between the subjects of Process and the subjects of State would be lost. Fortunately, I do not think that movement from Spec,State to Spec,Process is necessary. Notice that suffering a change of state entails being an undergoer. Consequently, the interpretation of the event does not require raising from Spec,State to Spec,Process because this movement does not add anything to the distribution of semantic roles. Since this movement is superfluous, I take it that it does not exist.
8 An anonymous reviewer points out that the noun head and the KP are not necessarily adjacent, since an adjective can stand between them.
el ataque brutal a la ciudad
the attack brutal DOM the city
‘the brutal attack on the city’
??el ataque a la ciudad brutal
Notice the position of the adjective in the equivalent English noun phrases:
the brutal attack on the city
*the attack brutal on the city
*the attack on the city brutal
The question is whether this should lead to the conclusion that there is in fact more structure in process nominals than I propose. The matter is complex, and a full treatment would take us too far afield. A provisional analysis that can be drawn from the contrast with English is that the Spanish n+√root raises to a higher position within the nominal complex, thus accounting for the lack of linear adjacency between n and the internal argument (this is probably on the right track although somewhat simplistic; see Cinque 2010 for a detailed investigation of word order in noun phrases).
9 Non-psych stative predicates do not accept n-DOM.
el peso de / *a 50 kilos
the weight GEN / DOM
50 kilos ‘the weight of 50 kilos’
el encuentro de la carretera con el río
the meeting GEN the road with the river
‘the meeting of the road with the river’
Ramchand (2008) places these predicates together with psych predicates in a single category, called rheme. It is unclear to me whether (i) and (ii) should have the same structure as (49a–c). An anonymous reviewer suggests that it may be the case that the arguments of stative predicates (including those in (51)) are not really internal but instead are related to the Voice structure. I regret having to postpone the matter for future research.
I would like to thank Julie Anne Legate for valuable and detailed comments on an earlier draft of this article. I would also like to thank Artemis Alexiadou and Karlos Arregi for discussion and the following audiences for their questions and feedback: Societas Linguisticae Europaea in Leiden; Hispanic Linguistics Symposium at the University of Illinois, Urbana-Champaign; University of Wuppertal; University of Göttingen. Special thanks go to the anonymous reviewers for Linguistic Inquiry, who went way beyond the call of duty to provide helpful discussion on my original submission. Finally, I would like to gratefully acknowledge the support of the Alexander von Humboldt Foundation in the form of a resumption fellowship as well as the hospitality of the Zentrum für Allgemeine Sprachwissenschaft in Berlin. I am the sole owner of all the mistakes that surely infect this article.