- Email: [email protected]
What's nominal in nominalizations?
Lingua 121 (2011) 1283–1296
Contents lists available at ScienceDirect
Lingua journal homepage: www.elsevier.com/locate/lingua
What’s nominal in nominalizations? Eric Reuland Utrecht Institute of Linguistics OTS, Utrecht University, Faculty of Humanities, Achter de Dom 20, 3512 JP Utrecht, The Netherlands
A R T I C L E I N F O
A B S T R A C T
Article history: Received 14 April 2008 Received in revised form 18 August 2009 Accepted 20 January 2011
This article investigates the nature of syntactic categories: What does it mean to be a noun or a verb? The discussion focuses on nominalizations. It is shown that there is a distinction between nouns and verbs at a basic level. The essential difference between verbs and nouns resides in the availability in principle of merging instructions in the sense of Reinhart (2002). What makes a nominalized verb nominal is consequently shown to reside in the instructions determining the way it is merged in the syntactic representation. Modifying a thesis formulated in Vinokurova (2005), it will be shown that Verbs represent relational concepts for which merging instructions are defined, and Nouns represent concepts for which merging instructions are not defined. ß 2011 Elsevier B.V. All rights reserved.
Keywords: Noun Verb Theta-role Merging instruction Nominalization Concept Lexicon
1. The problem One of the intriguing issues concerning the architecture of the language system is the relation between the conceptual system and the syntactic system (CHL in the sense of Chomsky 1995). One of the major questions in this area concerns the nature of syntactic categories. Specifically, the question has come up: What makes a lexical item nominal or verbal, etc.? We all know that languages have nominalizers applying to verbal elements and verbalizers applying to nominal elements. However, are categories undifferentiated at the level of the root? Is being a noun or a verb just a matter of arbitrary formal marking? Or are there concepts that are basically one or the other? And perhaps more concretely, what does it mean to be a noun or a verb? A useful window into this question is provided by nominalization structures. In what sense is a nominalized verb a noun (or a nominalized VP an NP/DP), and vice versa? What happens to its verbal properties? In this contribution, I will limit myself to the question of what is nominal in nominalizations, leaving the many other questions related to this issue to another occasion. I will show that there is a distinction between nouns and verbs at a basic level. What makes a nominalized verb nominal is shown to reside in the instructions determining the way it is merged in the syntactic representation. I will start my discussion with an analysis of how nominalization works in Dutch, and determine what it tells us. I then sketch the theoretical issues it raises followed by an overview of Reinhart’s (2002) theory of the relation between the system of concepts and syntactic structure, and draw some preliminary conclusions. I continue the discussion with nominalized constructions in English that parallel the Dutch examples, followed by a section presenting our results and some issues for further research.
E-mail address: [email protected]. 0024-3841/$ – see front matter ß 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.lingua.2011.01.015
1284
E. Reuland / Lingua 121 (2011) 1283–1296
2. Dutch nominal infinitives Like many languages Dutch has expressions showing a mix of nominal and verbal characteristics: the nominal infinitives. In Dutch they come in three types, reflected in the choice of determiner: 1. Het-determiner (the neuter form of the definite article) 2. Dat-determiner (the neuter form of the demonstrative) 3. No determiner Schoorlemmer (2001) discusses types 1. and 2., illustrated in (1a,b). (1c) illustrates type 3. (1)
a.
Het oproepen
van getuigen door de officier
the summon-INF of b.
the coroner
Dat afschuwelijke overlast veroorzaken van jou that terrible
c.
witnesses by
trouble cause-INF
of you
Bomen kappen (door de industrie) is schadelijk trees
felling (by the industry)
is harmful
Schoorlemmer notes a number of interesting contrasts between type 1 and type 2 infinitives, which I will not discuss here, and further presents an overview of nominal and verbal properties of nominal infinitives, as identified in the literature (Van Haaften et al., 1985; Hoekstra and Wehrmann, 1985; Reuland, 1988; Hoekstra, 1997). Nominal infinitives share with canonical DPs the following properties:
They are introduced by a determiner. They can be modified by adjectives. A subject or object may be expressed in a van-phrase (NB: there can be only one van-phrase). The van-phrase always occurs in a post-head position. The following properties are shared with VPs:
They can be modified by adverbs. Argument PPs can precede or follow the head. The subject can be realized as a door-phrase (by-phrase). (Bare accusative) direct objects must precede the verb. They may contain complex verbal structures. A number of properties deserve specific attention:
The direct object can be realized both preverbally as an NP or postverbally as a van-PP (of-PP). Modifiers can be realized with a schwa marking them as adjectives, or without a schwa marking them as adverbs. One and the same construction can be modified both by an adjective and an adverb, but only in that order. The presence of a determiner is necessary to license modification by an adjective (marked with a schwa). This entails that adjectives are impossible in type 3 nominalizations as illustrated in (2): (2)
a.
*FrequenteADJ/frequentADV bomen kappen (door de industrie) is schadelijk. frequent/frequently trees felling (by the industry) is harmful
b.
*FrequenteADJ/frequentADV kappen van bomen (door de industrie) is schadelijk. frequent/frequently felling of trees (by the industry) is harmful
Some further examples of type1 showing the main pattern are given in (3) and (4): (3)
a.
Het frequenteADJ/frequentADV bomen kappen (door de industrie) is schadelijk.
b.
*Het frequenteADJ/frequentADV kappen bomen (door de industrie) is schadelijk.
c.
Het frequenteADJ/frequentADV kappen (van bomen) (door de industrie) . . .
The frequent/frequently trees felling (by the industry) is harmful The frequent/frequently felling trees (by the industry) is harmful The frequent/frequently felling of trees (by the industry) is harmful
E. Reuland / Lingua 121 (2011) 1283–1296
d.
*Het frequenteADJ/frequentADV van bomen kappen (door de industrie) . . .
e.
Het *frequenteADJ/OKfrequentADV door de industrie bomen kappen is schadelijk.
1285
The frequent/frequently of trees felling (by the industry) is harmful The frequent/frequently by the industry trees felling is harmful f.
*Het frequenteADJ/frequentADV door de industrie kappen van bomen is schadelijk.
g.
Hun frequenteADJ/(?)frequentADV bomen kappen is schadelijk
The frequent/frequently by the industry felling of trees is harmful Their frequent/frequently trees felling is harmful h.
Hun frequenteADJ/frequentADV kappen van bomen is schadelijk Their frequent/frequently felling of trees is harmful
(4)
Hun frequenteADJ/*frequentADV kappen (door/*van de industrie) is schadelijk Their frequent/frequently felling (by the industry) is harmful (POSS is OBJ)
However: How verbal is verbal, how nominal is nominal? First of all, consider the question of how verbal is verbal in the light of the puzzle posed by Argument optionality. ‘Truly’ verbal and nominal complementation show the following contrast: A suitable discourse can license absence of the ‘obligatory’ complement in nominals, but not with true verbs. (5).
a.
De bomen moesten allemaal weg. Het vellen kostte veel tijd. The trees all had to go. The felling cost a lot of time.
b.
De bomen moesten allemaal weg.* Jan velde gedurende drie uur The trees all had to go. John fell for three hours.
(6)
a.
Dit struikgewas is schadelijk. Frequent verwijderen (door een expert) is nodig. This shrubbery is harmful. Frequent removing (by an expert) is needed.
b.
Dit struikgewas is schadelijk. *Een expert moet frequent verwijderen. This shrubbery is harmful. An expert should frequently remove.
(7)
a.
De stad viel in de handen van de vijand. Het vernietigen/de vernietiging kostte veel tijd. The town fell into the hands of the enemy. The destruction took a lot of time.
b.
De stad lag op de route van het vijandelijke leger. De vernietiging door de vijand kostte echter veel tijd./Het vernietigen door de vijand kostte echter veel tijd.
c.
De stad viel in de handen van de vijand.*De vijand was lang bezig te vernietigen.
a.
Emoties zijn zwaar. Frequent uitdrukken is belangrijk. Ongeremde uitdrukking is goed.
The town was in the way of the enemy army. The destruction by the enemy took a lot of time, however. The town fell into the hands of the enemy. The enemy took a long time destroying. (8)
Emotions are heavy. Frequent expression is important. Uninhibited expression is good. b.
Emoties zijn zwaar. *Jan drukte ongeremd uit omdat dat belangrijk was.
a.
Deze problemen zijn onoplosbaar. Veelvuldig opgeven moet vermeden worden.
Emotions are heavy. John uninhibitedly expressed because it was important. (9)
These problems are unsolvable. Frequent assignment should be avoided. b.
Deze problemen zijn onoplosbaar. *We geven desondanks constant op. These problems are unsolvable. Nevertheless we assign constantly.
These examples show that Dutch has considerable freedom in the realization of arguments in nominal infinitives, and that arguments that are syntactically obligatory in verbal constructions are not syntactically obligatory in the corresponding nominal infinitives. This is independent of the character of the modifier, whether or not a ‘subject’ is realized as a POSS phrase or a by-phrase, etc. This property obtains throughout in all other types of nominalization. The question is why this is so. A possible answer would be that these constructions are fully nominal and that nouns do not have obligatory arguments (contra Grimshaw, 1990, but see below for discussion of Grimshaw’s facts). However, this leads to the following further questions: (i) What does it mean to be nominal? (ii) Why would this property hold of nominals? These are the main questions I will start addressing now.
E. Reuland / Lingua 121 (2011) 1283–1296
1286
3. What does it mean to be a nominal? The question of what is means to be nominal is a central issue in the theory of syntactic categories. Giving an overview of the discussion of syntactic categories in the literature up to the present would lead me beyond the scope of this article (see Vinokurova, 2005 for such an overview and illuminating discussion). I will take as my starting point certain issues as they arise from the current framework of Distributive Morphology (Halle and Marantz, 1993; Marantz, 1997; Embick, 1997; Harley and Noyer, 1998; Harley and Noyer, 1999; Borer, 2003, etc.). This framework is based on the following tenets1: (10)
Lexical roots are category neutral, they are assigned a category X by merging with a small x. n+H=N a+H=A v+H=V
However, this tenet does not express what it means to be an n, rather than a v or a. It effectively states the following: - any root can merge with nominalizer to become a noun, a verbalizer to become a verb, an adjectivalizer to become an adjective. - there are no asymmetries between members or lexical categories. - all lexical categories are morphosyntactically complex. Vinokurova (2005) argues that in its generality, this view cannot be maintained. In particular, she shows it does not hold for Sakha, a Turkic—agglutinating—language spoken in the Sakha republic (Russia, East Siberia). For languages such as English with extensive 0-morphology it only apparently applies. Instead, she argues that a distinction exists between primitive verbs and primitive nouns. Consider cases of Noun–Verb pairs as in (11): (11)
a carpet – to carpet, an urchin – to urchin, a cut – to cut, a break – to break
In languages such as Sakha there is a direction of derivation as given in (12): (12)
a.
(N ! V) a carpet ! to carpet, an urchin ! to urchin;
b.
(V ! N) to cut ! a cut, to break ! a break.
In Sakha nouns such as those in (12a) and verbs as in (12b) are simplex. The corresponding verbs in (12a) and nouns in (12b) are morphologically complex involving ‘carpetnoun’, ‘urchinnoun’, ‘cutverb’, ‘breakverb’ plus some affixes. This is illustrated in (13): (13)
Noun member: structure
Verb member: structure
Vinokurova
N
N-Verbalizer
DM/XS
H-nominalizer
H-verbalizer
Example
Uot ‘fire’
Uot-taa ‘fire’
from
Bas ‘head, master’
Bah-yj ‘master’
Sakha
Baqa ‘desire’
Baqa-r ‘desire’
Vinokurova
V-Nominalizer
V
Example
Tu¨m-u¨k ‘conclusion’
Tu¨m ‘conclude’
from
Kepsee-n ‘story, narration’
Kepsee ‘tell, narrate’
Sakha
Taraa-x ‘comb’
Taraa ‘comb’
1 See for instance the following summary in Harley and Noyer (1999): Specifically, the different ‘parts of speech’ can be defined as a single l-morpheme, or Root (. . . .) in certain local relations with category defining f-morphemes. For example, a ‘noun’ or a ‘nominalization’ is a Root whose nearest c-commanding f-morpheme (or licenser) is a Determiner, a ‘verb’ is a Root whose nearest c-commanding f-morphemes are v, Aspect and Tense (. . . .). In more recent work, Borer gives a detailed analysis of the role of event structure in the mapping procedure, which I cannot possibly do justice to here. In essence her approach stays within the limits of what I sketch here.
E. Reuland / Lingua 121 (2011) 1283–1296
1287
The reverse derivations are simply impossible. What is needed is an independent characterization of what is nominal and verbal to explain the direction of derivation. An interesting proposal for such an independent characterization is given in Baker (2003). Baker proposes the following fundamental characteristic for nominals: Nominals satisfy Geach’s criterion of identity They can be used in same-contexts: this tree is the same as that one, etc. Syntactically this property is encoded as being carrier of an index. Baker acknowledges that indices violate inclusiveness. Therefore, he suggests that indices are added at the CI interface. This leads to the question of how syntax knows that a noun is a noun (again, independently from some category label, since it is the latter we wish to reduce). Some item x cannot be classified as a noun or a non-noun on the basis of inspection alone. Note, furthermore, that the criterion of identity is ill-suited to determine what is nominal in nominalizations, and certainly dubious in eventive expressions, as illustrated in (14): (14)
a.
? this walk is the same as that one
b.
*The felling of the trees by John is the same as the felling of the trees by Bill
If so, this leads us once more to the question of what it is that is nominal in nominalizations. (See Baker, 2003 for an overview of problems in using other criteria, e.g. morphological criteria like types of affixation.) What we need is a criterion whose applicability can be determined by inspection. Vinokurova (2005) presents the following important insight: - verbs are relational, nouns are not relational, adjectives are properties Since prima facie not all verbs are relational (witness passives and unaccusatives), it is necessary to make this notion precise. To do so, I will discuss the implementation of this idea within the theory of thematic roles presented in the Thetasystem developed by Reinhart (2002), Reinhart and Siloni (2005), Marelj (2004), and others. As we will see, one of the core ideas of that theory is the active lexicon, wherein there are lexical operations that may derive unary predicates from underlyingly relational concepts. 4. The Theta-system: the central system of concepts I will start with a concise overview of the main properties of the Theta-system. Its basic tenets are the following: (i) Abstract concepts need to be formally coded in order to be legible to the Computational System (CS); (ii) The Theta-system enables the interface between the system of concepts and the CS as well as the interface between the system of concepts and the inference system. The Theta-system is the narrow-channel by which the computational system can access information from the conceptual system. Whatever conceptual information cannot pass through this channel cannot be input to the syntactic computation. The Theta-system comprises the components in (16): (16)
Components of the Theta-system a.
A system for formally encoding the information about abstract concepts the computational system must be able to access: the linguistic coding of causal relationships;
b.
Lexical entries (i.e. particular coded concepts);
c.
Operations on entries;
d.
Merging instructions (linking/mapping rules).
I will briefly sketch these in turn. 4.1. The system The conceptual information the computational system can see is in the form of theta-roles. All theta roles are formally coded as clusters defined in terms of two binary features: [c] (cause change), [m] (mental state).
E. Reuland / Lingua 121 (2011) 1283–1296
1288
A cluster need not be specified for (all) features. This allows the null-cluster as a limiting case.2 The features are characterized in (17): (17)
The features a.
The c-feature The feature/+c represents that the role is perceived as a sufficient condition (Shen, 1985). The feature/c represents that the argument is identified as not being the cause. If the c-feature is not represented, the causal status of the argument is undetermined.
b.
The m-feature The/+m feature is related to the role of the mental state of the participant present in the relation ‘motivate’. Hence, the/+m feature entails animacy. A/m specification does not entail in-animacy.
4.2. Lexical entries The thematic properties of Verbs (i.e. n-place predicates) are encoded in terms of these feature clusters. The following definitions apply: (18)
Definitions and notations [α]
= Feature cluster α.
/α
= Feature (and value) α. (E.g. the feature/+m occurs in the clusters [+c + m], [-c + m], and [+m])
[/α]
= A cluster one of whose features is/α. (E.g. [/-c] clusters are [-c + m], [-c-m] and [-c].)
[+]
= A cluster all of whose features have the value +.
[]
= A cluster all of whose features have the value -.
These definitions are necessary to formulate well-formedness conditions on lexical entries, such as the Cluster Distinctness Constraint (CDC), which states that two indistinct theta-clusters cannot be both realized on the same predicate. (The CDC rules out realizations of a verb like anger as in The article angered Bill at the government, see Reinhart (2002) for more discussion). Importantly for our present purposes, they are needed to specify the lexical operations discussed in section 3.3, and the linking conditions to be discussed in section 3.4. The Theta-system specifies a restricted number of operations on lexical entries expressing significant relations between lexical items. This part of the theory will be sketched next. 4.3. Operations on lexical entries The lexical operations in the Theta-system are tied to the following principle: (19)
Lexicon Uniformity Hypothesis Each verb-concept corresponds to one lexical entry with one thematic structure. The various thematic forms of a given verb are derived by lexicon operations from one thematic structure.
This principle entails that the unaccusative forms in (20b) and (21b) are derived from the basic transitive forms in (20a) and (21a). (20)
2
a.
Peter melted the ice.
b.
The ice melted.
The two binary features define nine feature clusters:
(1)
[+c+m]
Agent
(2)
[+c-m]
Instrument
(3)
[-c+m]
Experiencer
(4)
[-c-m]
Theme
(5)
[+c]
Cause
(6)
[-c]
Recipient Goal/Benefactor
(7)
[-m]
Subject Matter/Source
(8)
[+m]
Sentient
(9)
[-]
Arb in middles
E. Reuland / Lingua 121 (2011) 1283–1296
(21)
a.
Peter froze the ice.
b.
The ice froze
1289
Reinhart assumes that unaccusatives are lexically derived also when the basic transitive form is not realized in a language.3 The available operations on verbal concepts are: i. Reduction ii. Saturation (e.g. Passive) iii. Entry-changing (e.g. Lexical Causativization) To give the reader some background about the way the system works, I summarize its main operations. This is important to enable the reader to assess what is involved in the nominal-verbal contrast. i. Reduction Reduction affects the number of positions in the grid of a lexical element that are linked to an argument bearing a thematic role. The way it operates depends on whether an internal or the external role is affected. (i) Reflexive (unergative) entries are the output of reduction of the internal role. (ii) Unaccusative entries are the output of reduction of the external role. The two are not entirely symmetrical. When the internal role is reduced, it retains its semantic effect. The result is a composite role rather than just the original external role, as illustrated in (22). (22)
Internal Reduction/Bundling: a. Vacc (θ1, θ2) ! Rs(V) (θ1,2) b. Rs (V)(θ1,2) $ θ1,2 (λx (V (x,x))) c. shaveacc([+c + m]1,[-c-m]2: Lucie shaved him. d. Rs(shave)([+c + m]1): Max shaved.
When the external role is reduced, it is no longer semantically visible, as we see in (23): (23)
Expletivization: Reduction of an external [+c] role (semantically null function) a. Vacc (θ 1[+c], θ2) ! Re V(θ2) b. Re(V) (θ2) = V(θ2) c. openacc ([+c], [-c-m] ! Re(open)[-c-m] d. worryacc ([+c], [-c + m] ! Re(worry)[-c + m]
The following conditions on the availability of these operations should be noted: (i) Reduction can apply only in a two-(or more)-place entry, one of whose roles is external (a [+] cluster). (ii) Reduction eliminates the accusative feature of the verb (fully or partially, see for discussion below). (iii) Reduction and saturation (passive) are exclusive (only one can apply at each entry). Consider next the operation of saturation. ii. Saturation A crucial component of saturation is variable binding: A variable in the representation of the conceptual structure is bound by some operator, e.g. an existential or generic operator (E or Gen). Saturation entails that the theta-role is unavailable for syntactic purposes. An example is passivization. Passive saturation closes existentially the external argument of a verb as illustrated in (24): (24)
a. b.
wash (θ1, θ2) Ex (wash (x, θ2)) Max was washed
! Ex (x washed Max)
In addition to reduction and saturation, there are also entry-changing operations. iii. Entry-Changing Operations An example is lexical causativization, illustrated in (25) and (26): (25) (26)
a. b.
The dog walked ! She walked the dog. The horse jumped ! The rider jumped the horse over the fence
The components of this operation are given in (27). It involves both changing the content of a cluster and adding a cluster: 3
I will come back to this below.
E. Reuland / Lingua 121 (2011) 1283–1296
1290
(27)
Lexical Causativization a.
Decausativize: Change a/ + c feature to a/-c feature.
b.
Agentivize: Add an agent role.
walk: ([+c + m]) ! walk: ([-c + m]) walk: ([-c + m]) ! walk: ([+c + m], [-c + m]) Languages may differ as to where such operations take place. This variation is expressed in the Lex-Syn Parameter: UG allows thematic arity operations to apply in the lexicon or in the syntax. Concept forming operations can only take place in the lexicon. Some further variation takes place in the instructions for merging and linking arguments to the theta-roles projected by the lexical element. 4.4. Merging/linking instructions Linking instructions are stated in terms of s-selection (i.e. the type of arguments that are encoded on the verb’s grid) and the following set of rules and generalizations: (28)
Lexicon marking Given an n-place verb-entry, n > 1, a.
(29)
(30)
Mark a [] cluster with index 2.4
b.
Mark a [+] cluster with index 1.
c.
If the entry includes both a [+] cluster and a fully specified cluster [/α,/-c], mark the verb with the ACC feature.
Relevant generalizations of lexical operations: a.
Saturation and reduction apply to the marked entry (i.e. after marking).
b.
Reduction eliminates the accusative feature of the verb (fully or partially).
CS merging instructions. a.
When nothing rules this out, merge externally.
b.
An argument realizing a cluster marked 2 merges internally; an argument with a cluster marked 1 merges externally.
The way reduction affects Case reflects another source of cross-linguistic variation. As we saw in (22), repeated here, in English the ACC feature is fully eliminated by reflexive reduction: (22)
Internal Reduction/Bundling: a.
Vacc (θ1, θ2) ! Rs(V) (θ1,2)
b.
Rs (V)(θ1,2) $ θ1,2 (λx (V (x,x)))
c.
shaveacc([+c + m]1,[-c-m]2: Lucie shaved him.
d.
Rs(shave)([+c + m]1): Max shaved.
The object position of shave in (22d) is just empty. Dutch and many other languages do not allow this position to be empty. Rather, a simplex anaphor is required, as in the Dutch equivalent of (22d) Max scheert zich. Here, the simplex anaphor zich must be inserted to check a residual structural Case. Another way of saying this is that in Dutch, the reflexive entry of scheren is still syntactically 2-place. The internal role reduction only eliminates part of the Case frame, the thematic part is eliminated, but the structural Case component is still there and requires checking.5 For sake of concreteness, I give a sample derivation of a transitive structure in (31) and an unaccusative derivation in (32). (31)
Transitive Derivations a.
Max ate an apple.
b.
Base entry: eat ([+c + m], [-c-m])
4 Note that these indices are just annotations for the merging instructions to apply. They should not be confused with referential indices, which the inclusiveness condition prohibits. 5 This presupposes a componential analysis of ACC case, into structural and thematic. See Belletti (1988) for an earlier version of this idea. For extensive discussion from the current perspective, see Reuland (2011).
E. Reuland / Lingua 121 (2011) 1283–1296
(32)
c.
Marking: eatacc ([+c + m]1, [-c-m]2)
d.
Merging: By (30b), ([+c + m]1) merges externally, ([-c-m]2 merges internally
1291
Unaccusative Derivations a.
The vase broke.
b.
Base entry: break ([+c], [-c-m])
c.
Marking: breakacc ([+c]1, [-c-m]2)
d.
Expletivization applies to the marked entry, giving rise to (break ([-c-m]2))
e.
Merging: Internal, by (30b)
Just as in the case of internal reduction/reflexivization, languages may vary as to whether ACC is eliminated by the reduction operation, or whether a Case residue remains. If a case residue remains, it has to be checked. Checking can take place by a simplex reflexive element, like Dutch zich, by a clitic, or some other marker. It is not always straightforward to determine why precisely a marker has to be present, and what it contributes. In German we see the element sich with middles, reflexives and unaccusatives. In Italian we see the clitic si again with middles, reflexives and unaccusatives, but also with passives. The same properties are found with the reflexive clitic se in SerboCroation. The question is then what unifies such seemingly disparate uses. What these cases share is that an operation on argument structure has applied that is licensed by sich/si/se. In Sakha we see a similar pattern. As discussed extensively in Vinokurova (2005), Sakha has a range of reflexive, passive and also a yet wider range of unaccusative entries that require the presence of a special affix –n. For sake of illustration, (33) gives some causative/unaccusative pairs from Vinokurova (2005, App. IV. II): (33)
Causative
Causative Gloss
Unaccusative
Tiree
Prop, support, lean
Tiren
Aqaa
Smear, glue
Aqan
Tox
Spill
Toqun
From a synchronic perspective, Sakha is best described as a language lacking structural Case. Vinokurova concludes that the most straightforward generalization is that –n in Sakha marks the application of a (reduction or saturation) operation on argument structure. How, precisely, it contributes to licensing would be an interesting topic for further research.6 For current purposes it suffices that Sakha clearly shows morpho-syntactic effects of reduction in a wide range of unaccusative entries. 5. Elaboration 5.1. 1-place verbs Coming back to our main concerns, as we saw, the Theta-system derives unaccusative entries from 2-place entries. Vinokurova (2005) now proposes to generalize this property. That is, what sets apart basic verbs from basic nouns is that all true verbs have a 2-place source, and basic nouns do not. This tallies with the analysis of unergative verbs, the other type of prima facie 1-place verbal predicates, put forward in Hale and Keyser (1993, 1998, 2000). These authors analyze unergative verbs as denominal: transitive structures involving a verbalizer and a nominal complement. Following this line, Vinokurova proposes that yet another class, namely weather-verbs, also be analyzed as unergatives.7 This leads to what I will refer to as Vinokurova’s thesis: 6 The proposal that unaccusatives are systematically derived from two-place entries by a reduction operation whose effects show up in particular markings is orthogonal to the morpho-syntactic manifestation of transitivity. Leon Stassen (personal communication) mentions that in some languages transitives appear to be derived from 1-place verbs in the sense that they are morphologically more complex. As always, languages do not bear their analysis on their sleeves. What would be needed for firm conclusions is an in-depth investigation of the nature and distribution of such morphemes. Whatever complexity shows up could well be the reflection of a Case frame, or object agreement. In principle, languages could exist in which a substantial class of ‘ordinary’ transitive concepts have obligatory external argument reduction, with only unaccusatives surfacing, an inventory that would then be supplemented by the operation of an entry-changing causativizing affix as discussed above. Mutaka and Safir (2008) (M&S) describe a very intriguing phenomenon in Kinande. Kinande, -nab- ‘‘wash’’ and -hakab- ‘‘smear’’ have intransitive reflexive interpretations like their counterparts in many other languages. As in English there is a corresponding transitive reflexive form. In English, the transitive reflexive is formed by using a self-form (himself, etc.). In Kinande the productive strategy is not available, unless the verb stem is extended with the morpheme –i. The question is how this fact is to be interpreted. M&S argue that this indicates a direction of derivation opposite to what Reinhart and Siloni claim for English. M&S discuss a number of interesting further properties of this morpheme, without reaching a definitive conclusion about its properties. As they say, their approach to -i- is somewhere between those that treat its Bantu cognates simply as causative or direct causative (. . . ) and those that treat it as strictly a transitivizer. Unlike what they suggest, I do not think this result is necessarily a problem for Reinhart & Siloni’s approach. If the i-affix is a causative verb extension, such a fact is compatible, as we saw. Moreover, we know independently that many languages do mark unaccusatives. Thus, taking all the facts together the fundamental question appears to be what would motivate a see-saw effect in Kinande and similar languages. Such an investigation would lead us beyond the scope of the present contribution, however. 7 Vinokurova does not discuss the status of theme-unergatives, which include emission verbs such as glow. It would seem that they are amenable to the same analysis.
1292
(34)
E. Reuland / Lingua 121 (2011) 1283–1296
Vinokurova’s thesis: Verbs represent intrinsically relational concepts in opposition to verbs: noun concepts are intrinsically non-relational Adjectives are intrinsically 1-place concepts
5.2. (Relational) nouns If noun concepts are intrinsically non-relational, what about nouns such as sister and father? These are generally analyzed as relational: a sister is necessarily somebody’s sister, a father somebody’s father. Vinokurova argues that the relevant notion of being relational is made explicit by the properties of theta-clusters: A verbal concept like EAT is associated with two necessary conditions – someone to perform the eating process (also a sufficient condition) and something to be eaten. These causal relations are translated into θ-features yielding two arguments of the verb eat. With relational nouns like sister, on the contrary, it is not motivated in terms of causality that they should take an argument. This is because a necessary and sufficient condition for being a sister is being a female sibling. However, the concepts FEMALE SIBLING and SISTER are coextensive. Therefore, with relational nouns it is impossible to translate their causal relations into distinct feature clusters – the potential arguments. (Vinokurova, 2005:77). Hence, basic nouns do not have a theta-grid in the relevant sense. This entails that they cannot undergo lexical operations defined on theta-grids. 6. How to proceed from here? This being said, we still have to address our main problem: What is nominal in nominalizations? I will briefly discuss three options. Option 1: The status of an element as nominal or verbal can be determined by inspection of the concept. This entails that the categories V, N and A become predictable, hence redundant. Functional categories select for a lexical item with a relational conceptual structure (!V), a non-relational conceptual structure (! N), or a 1-place conceptual structure (! A). Since the arity reducing operation of Reduction/Bundling applies to the marked concept, derived 1-place verb-concepts reflect that they are derived from a relational verb/concept. However, if this is all, the status of nominalizations, which somehow combine nominal and verbal characteristics, is still left open. Let’s therefore consider a further option. Option 2: Reconsider one part of the procedure in the theta-system. According to the original version of the theta-system, the argument structure of verbal concepts is made readable for CHL by channelling it through the [c], [m] system. It assumes that being verbal is given. Consider now the following addition: Being channelled through the feature system is what makes a concept verbal. The [ c], [m] system articulates the concept. If a concept is not relational, channelling it through the [c], [m] system can either be assumed to apply vacuously, or be considered as undefined. In any case the concept does not become relational. If a concept is relational it can, but need not be articulated by the [c], [m] system. If yes, it is realized as verbal. If not, it is realized as nominal. If so, categorization as V, N and A is again predictable. It can be read off, and need not be annotated. Consider, for instance, the concept vellen ‘fell’. It is potentially relational. If channelled though the [c], [m] system, its theta-clusters will be articulated, and it will be realized as a verb. Having articulated theta-clusters is visible by inspection, and its theta-clusters will be subject to the merging procedure. If not channelled through the [c], [m] system, it has no clusters that are subject to the merging procedure; hence, its arguments are realized optionally. A potential problem with this line is that one would expect that ‘arguments’ of nouns/nominalizations may connect (in some way or other) with any meaning aspect of the concept. Arguments of nouns would be able to bear ‘roles’ that cannot be encoded in the [c], [m] system, hence, are unlike any theta-roles of verbal arguments. There is no evidence that this obtains. Let us therefore consider a yet more restrictive option: Option 3: The essential difference between verbs and nouns resides in the availability in principle of merging instructions. The roles that nouns can assign (if they have the right conceptual make-up) are just the roles that verbs can assign; however, they are not accessible to marking in the sense of (28). This entails that (30) is undefined and leads the following modification of Vinokurova’s thesis:
E. Reuland / Lingua 121 (2011) 1283–1296
(35)
1293
Vinokurova’s thesis modified:
Verbs represent relational concepts for which merging instructions are defined
Nouns represent concepts for which merging instructions are not defined
(Suggestion: Adjectives represent concepts with merging instructions for one argument)
In this form it is applicable both to basic nouns and to ‘nominalized’ verbs. Selection still holds in relation to nominals, which entails that arguments can be realized using appropriate prepositions, and is consistent with Grimshaw’s (1990) thesis that nouns are deficient theta-markers. (35) has the following consequence: If no merging instructions for the complements of a lexical item are defined, its complements cannot be obligatory. As an anonymous reviewer notes, nouns do mark their arguments, for instance, the direct object by the preposition of in the destruction of Rome, or by the genitive as in Rome’s destruction, or the Vandal’s destruction of Rome. Thus, one may wonder whether nouns would not have merging instructions after all. Interestingly, Higginbotham (1983) shows convincingly that the element in the POSS position does not bear a theta role determined by the noun in the same way as a subject bears a theta-role determined by the verb. Rather he argues, the role it receives in interpretation is determined by a general aboutness-relation, as for instance illustrated by the use of expressions like yesterday in yesterday’s examination. Just like no merging instruction of the type specified in (28–30) says that yesterday has to be ‘‘merged externally’’ to realize a particular theta cluster, no merging instruction specifies that the Vandals would have to merge externally in order to realize a particular cluster in The Vandal’s destruction of Rome, and in fact, it can equally well express agenthood in a by-phrase as in The destruction of Rome by the Vandals. Thus, clearly, the concept destroy allows for an agent. An agent relation can be picked up in the POSS position, by virtue of Higginbotham’s aboutness strategy, or in a position where this strategy is not available on the basis of a preposition marking the role. The same applies to the theme. Conceptually, it is available, but it is not associated with a particular merging instruction. For an argument in POSS position it is available by aboutness; for another position it must be explicitly marked by the preposition of. As noted above, this is in line with Grimshaw’s deficiency thesis. We now have an independent characterization of what’s nominal in Dutch nominal infinitives. Do we also have an explanation for the optionality of their arguments? I am grateful to an anonymous reviewer for bringing this question up. Clearly, there are different levels of explanation. Let us see how far we have come. Blindly applying the Theta-system we can say that it has been explicitly developed as a theory about verbal concepts, crucially relying on an articulation in terms of the features [c], [m]. We have an independent characterization of the representation of nominal structures as not being articulated in terms of these two features. As they are formulated, there is no way the merging instructions (28–30) could apply to such nominal structures. There is every reason to think that they do not, so we have an explanation at the level of a particular theory. Of course, it is always extremely useful to move beyond this and ask ourselves deeper why-questions. In most cases, and to my knowledge in this case as well, this requires moving beyond what we have explicit theories of, hence coming closer to speculation. But let’s give it a try. The core idea behind Reinhart’s feature system is that causation and mental involvement are basic categories in which we organize the world. Note the term organize. This organization is not inherent in reality, but in the way we may choose to represent reality, but are not obliged to. Thus, a destruction or examination or felling event can be represented in different ways. However, once we choose to do so in terms of causation and mental involvement the representation becomes necessarily relational, either in terms of our relation to the eventuality (mental involvement) or in terms of the internal organization of the eventuality (causation). This leads us back to Vinokurova’s core intuition: the relational character of verbs, versus the non-relational character of nouns. Of course, there are further questions to ask. Why do verbs – abstracting away from big V or little v’s – check structural accusative, and why is it that nouns do not? Or to put it differently, why are structural Cases (nominative/accusative) in some broad sense related to the Tens/aspect system? What is the relation between the dependent character of nominativeaccusative (see Marantz, 2000) and the eventive/relation character of verbs? These are all important questions, but any attempt at discussion would carry us beyond the scope of this contribution. Let us go back to Vinokurova’s thesis as we revised it in (35), and the way it accounts for the Dutch facts we started out with. In this form, it contradicts Grimshaw’s (1990) analysis of complex event nominals. That’s why the last section is devoted to this issue.
7. Complex event nominals in English8 Grimshaw (1990) argues that complex event nominals have argument structure, just like verbs. She claims that complements to complex event nominals are obligatory as in the case of the corresponding verb. The view that complements 8
I would like to thank Anna Asbury and Bill Philip for their help.
1294
E. Reuland / Lingua 121 (2011) 1283–1296
in nominalizations are not obligatory is due to the fact that many derived nominals are ambiguous between the complex event reading and a simple event/result reading. In her view, Given a proper disambiguation, the obligatoriness can be observed. The main difference between complex event nominals and verbs: Nominals are deficient theta-role assigners and require help from a preposition. Grimshaw presents the following examples of unambiguously complex event nominals and the obligatoriness of their arguments: (36)
a.
The felling *(of the trees)
b.
They felled *(trees)
c.
The destroying *(of the city)
d.
They destroyed *(the city)
Disambiguation by a modifier is shown in the following examples: (37)
(38)
a.
The expression is desirable
b.
*The frequent expression is desirable
c.
The frequent expression of one’s feelings is desirable
d.
We express *(our feelings)
a.
The assignment is to be avoided
b.
*The constant assignment is to be avoided
c.
The constant assignment of unsolvable problems is to be avoided
d.
We constantly assign *(unsolvable problems)
The effect of disambiguation by agents is shown in (39) and (40): (39)
a.
*The instructor’s intentional/deliberate examination took a long time.
b.
The instructor’s intentional/deliberate examination of the papers took a long time.
(40)
a.
The destruction was awful to see
b.
*The enemy’s destruction was awful to watch
c.
The enemy’s destruction of the city was awful to watch
d.
The enemy destroyed *(the city)
These facts appear to be unexpected given the Dutch facts from section 1. This leads to the question of how different English is from Dutch. However, properly considered, English complex event nominals are not very different at all. In fact, according to my informants, they behave just like Dutch. In (36)’–(40)’ I use English examples taken from the list above in the same context I used in the Dutch cases. The effect is striking: (36)’ (37)’ (38)’ (39)’ (40)’
a.
There were a lot of trees on the building lot. The felling took several days.
b.
There were a lot of trees on the building lot. *The workers felled for several days
a.
Suppressed emotions are dangerous. Frequent expression is desirable.
b.
Suppressed emotions are dangerous. *Therefore we expressed frequently
a.
These problems are unsolvable for these students. Regular assignment is to be avoided
b.
These problems are unsolvable for these students. *Therefore, we don’t assign regularly.
a.
The papers were extremely tough reading. The instructor’s careful examination took a long time.
b.
The papers were extremely tough reading. *The instructor carefully examined for a long time.
a.
The city was in the way of a strong enemy army. Yet, the complete destruction by the enemy took several days.
b.
The city was in the way of a strong enemy army. *The enemy completely destroyed, which took them several days.
A clear contrast emerges between arguments of complex event nominals and arguments of verbs: given an appropriate discourse, arguments that are obligatory for verbs can be omitted in nominalizations. Hence, English complex event
E. Reuland / Lingua 121 (2011) 1283–1296
1295
nominals are as compatible with Vinokurova’s modified thesis as their Dutch counterparts. This leads us to the question of what types of variation between nominals one may expect, as expressed in the title of the next section.9 8. Expected types of variation between nominal(ization)s I will be content giving just a list of types, with their characteristics. Although, no doubt, each would merit a more extensive discussion than I can give here, for present purposes this overview will suffice: 1. Basic nouns No internal relational structure ! no merging instructions ! nominal 2. Simple event nominals An affix selects a relational concept without articulated theta-structure ! no merging instructions ! nominal 3. Complex event nominals An affix selects a relational concept with articulated theta-structure ! blocks assignment of merging instructions ! nominal 4. Nominal infinitives An affix selects a relational concept with articulated theta-structure and (thematic) ACC case, and blocks assignment of merging instructions ! nominal 5. Gerunds A ‘‘nominalizing affix’’ that applies in the syntax leaving a full verb with merging instructions in its domain ! it is inflectional rather than derivational (gerundival –ing, etc.) ! not nominal Instantiates a class of elements licensing verbal projections to appear as arguments, functionally similar to complementizers. This overview leads to the following further questions: (i)
(ii)
a.
Do we have nominalization as a lexical operation on items that are fully verbal in the sense that they have explicit merging instructions for their arguments and that ‘takes away’ the merging instructions?
b.
Do we have nominalization as a syntactic operation on items that are fully verbal in the sense that they have explicit merging instructions for their arguments and that ‘takes away’ the merging instructions?
How precisely do nominalizers affect the item they merge with? I see the following options:
(i)
An affix may suppress (using Grimshaw’s term) an argument. This would entail that the concept becomes nonrelational and that merging instructions are not applicable. This requires that suppression applies, before lexicon marking (28) can apply. However, this can only account for some of the above types.
(ii)
Alternatively, the affix projects a structure to which the merging instructions cannot be applied. Viability of this option depends on implementation of the merging instructions.
Both require further research for a proper assessment. However, these relatively open questions notwithstanding, I feel we have come a long way towards an understanding of the question we started out with, which leads me to a brief final section. 9. Conclusion I would like to conclude saying that Vinokurova’s thesis modified explains the following: What is nominal in nominalizations Why arguments are optional in nominalizations Which is what we set out to do. 9 An anonymous reviewer notes that the data are "not as cut and dried" as they are presented here. The reviewer acknowledges that there is a contrast in the direction I found, but in many cases (s)he finds the sentences with argument omission more questionable than my informants. So, apparently there is idiolectal variation here that merits further investigation. Another interesting fact the reviewer brings up is that " –ing-nominals lacking overt arguments are virtually as bad as sentences even with contextual information provided", as in (i):
(i)
Repressed emotions are generally considered to be unhealthy. Most experts therefore recommend frequent expressing *(of them/of such emotions)
The interesting thing is that here the reviewer sees a clear contrast between the sentence with the –ing form expressing and the one with expression. It is well-known that certain English –ing-forms are fully verbal (Reuland, 1983). The reviewer’s observation suggests that for some English speakers the –ing forms are predominantly construed as containing a verbal projecting (with a PRO subject) also when embedded in a functional shell with DP properties licensing the adjectival form of the modifier and the of-form of the objective Case (except for (36a’) which the reviewer finds fully acceptable), that is as gerunds, which are listed as verbal in section 7.5 rather than as nominalizations. Clearly, there is a form of competition between construing an –ing-form as a gerund or as a nominalization if their further properties leave this open. What factor determines whether an –ing-form is construed as one or the other is again a matter for further investigation.
1296
E. Reuland / Lingua 121 (2011) 1283–1296
Acknowledgements Many thanks to Jaklin Kornfilt and John Whitman for organizing a very inspiring workshop and inviting me to participate and present my ideas. I am also indebted to the audience for their great talks, and stimulating discussion. Last but not least, I am very grateful to two anonymous reviewers for their very useful comments that helped me improve the final article both in form and content, and again to Jaklin and John for all their work in getting this collection published. Needless to say that I am solely responsible for all errors. References Baker, M., 2003. Lexical Categories: Verbs, Nouns, and Adjectives. Cambridge University Press, Cambridge. Belletti, A., 1988. The case of unaccusatives. Linguistic Inquiry 19, 1–34. Borer, H., 2003. Exo-skeletal vs. endo-skeletal explanations: syntactic projections and the lexicon. In: Moore, J., Polinsky, M. (Eds.), The Nature of Explanation in Linguistic Theory. CSLI Publications, Stanford, CA, pp. 31–67. Embick, D., 1997. Voice and the Interfaces of Syntax. Doctoral Dissertation, University of Pennsylvania, Philadelphia. Grimshaw, J., 1990. Argument Structure. MIT Press, Cambridge, MA. Haaften, T., et al., 1985. Nominalisaties in het Nederlands. Glot 8, 67–104. Hale, K., Keyser, S.J., 1993. On argument structure and the lexical expression of syntactic relations. In: Hale, K., Keyser, S.J. (Eds.), The View from Building 20, MIT Press, Cambridge, MA, pp. 53–109. Hale, K., Keyser, S.J., 1998. Bound features, Merge and transitivity alternations. Ms. MIT Press, Cambridge, MA. Hale, K., Keyser, S.J., 2000. Conflation. In: Cuadernos de Linguistica del Instituto Universitaria Ortega y Gasset 7, pp. 39–76. Halle, M., Marantz, A., 1993. Distributed morphology and the pieces of inflection. In: Hale, K., Keyser, S.J. (Eds.), The View from Building 20. MIT Press, Cambridge, MA, pp. 111–176. Harley, H., Noyer, R., 1998. Mixed nominalizations, short verb movement, and object shift in English. In: Tamanji, P.N., Kiyomo, K. (Eds.), Proceedings of NELS 28. GLSA, University of Massachusetts, Amherst, pp. 143–157. Harley, H., Noyer, R., 1999. State-of-the-article: distributed morphology. Glot 4 (4), 3–9. Higginbotham, J., 1983. Logical form, binding and nominals. Linguistic Inquiry 14, 395–420. Hoekstra, J., 1997. The Syntax of Infinitives in Frisian. Doctoral Dissertation, University of Groningen, Fryske Academy, Ljouwert. Hoekstra, T., Wehrmann, P., 1985. De nominale infinitief. Glot 8, 257–275. Marantz, A., 1997. No escape from syntax: don’t try morphological analysis in the privacy of your own Lexicon. In: Dimitriadis, A., et al. (Eds.), Proceedings of the 21st Annual Penn Linguistics Colloquium. Penn Working Papers in Linguistics 4 (2), pp. 201–225. Marantz, A., 2000. Case and licensing. In: Reuland, E. (Ed.), Arguments and Case: Explaining Burzio’s Generalization. John Benjamins, Amsterdam. Marelj, M., 2004. Middles and Argument Structure across Languages. Doctoral Dissertation. LOT series, Utrecht University. Mutaka, N.M., Safir, K., 2008. Kinande anaphora sketch. www.africananaphora.rutgers.edu/downloads/./KinandeAS.pdf. Reinhart, T., 2002. The Theta system – an overview. Theoretical Linguistics 28, 3. Reinhart, T., Siloni, T., 2005. The lexicon-syntax parameter: reflexivization and other arity operations. Linguistic Inquiry 389–436. Reuland, E., 1983. Governing –ing. Linguistic Inquiry 14 (1), 101–136. Reuland, E., 1988. Relating morphological and syntactic structure. In: Everaert, M., et al. (Eds.), Morphology and Modularity. Foris Publications, Dordrecht. Reuland, E., 2011. Anaphora and Language Design. MIT Press, Cambridge, MA. Schoorlemmer, M., 2001. Dutch nominalised infinitives as non-identical twins. Ms. UiL OTS, Utrecht University. Shen, Y., 1985. The Structure of Action in the Short Narrative Text. Unpublished PhD Dissertation, Tel Aviv University. Vinokurova, N., 2005. Lexical Categories and Argument Structure: A Study with Reference to Sakha. Doctoral Dissertation. LOT series, Utrecht University.
Contents lists available at ScienceDirect
Lingua journal homepage: www.elsevier.com/locate/lingua
What’s nominal in nominalizations? Eric Reuland Utrecht Institute of Linguistics OTS, Utrecht University, Faculty of Humanities, Achter de Dom 20, 3512 JP Utrecht, The Netherlands
A R T I C L E I N F O
A B S T R A C T
Article history: Received 14 April 2008 Received in revised form 18 August 2009 Accepted 20 January 2011
This article investigates the nature of syntactic categories: What does it mean to be a noun or a verb? The discussion focuses on nominalizations. It is shown that there is a distinction between nouns and verbs at a basic level. The essential difference between verbs and nouns resides in the availability in principle of merging instructions in the sense of Reinhart (2002). What makes a nominalized verb nominal is consequently shown to reside in the instructions determining the way it is merged in the syntactic representation. Modifying a thesis formulated in Vinokurova (2005), it will be shown that Verbs represent relational concepts for which merging instructions are defined, and Nouns represent concepts for which merging instructions are not defined. ß 2011 Elsevier B.V. All rights reserved.
Keywords: Noun Verb Theta-role Merging instruction Nominalization Concept Lexicon
1. The problem One of the intriguing issues concerning the architecture of the language system is the relation between the conceptual system and the syntactic system (CHL in the sense of Chomsky 1995). One of the major questions in this area concerns the nature of syntactic categories. Specifically, the question has come up: What makes a lexical item nominal or verbal, etc.? We all know that languages have nominalizers applying to verbal elements and verbalizers applying to nominal elements. However, are categories undifferentiated at the level of the root? Is being a noun or a verb just a matter of arbitrary formal marking? Or are there concepts that are basically one or the other? And perhaps more concretely, what does it mean to be a noun or a verb? A useful window into this question is provided by nominalization structures. In what sense is a nominalized verb a noun (or a nominalized VP an NP/DP), and vice versa? What happens to its verbal properties? In this contribution, I will limit myself to the question of what is nominal in nominalizations, leaving the many other questions related to this issue to another occasion. I will show that there is a distinction between nouns and verbs at a basic level. What makes a nominalized verb nominal is shown to reside in the instructions determining the way it is merged in the syntactic representation. I will start my discussion with an analysis of how nominalization works in Dutch, and determine what it tells us. I then sketch the theoretical issues it raises followed by an overview of Reinhart’s (2002) theory of the relation between the system of concepts and syntactic structure, and draw some preliminary conclusions. I continue the discussion with nominalized constructions in English that parallel the Dutch examples, followed by a section presenting our results and some issues for further research.
E-mail address: [email protected]. 0024-3841/$ – see front matter ß 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.lingua.2011.01.015
1284
E. Reuland / Lingua 121 (2011) 1283–1296
2. Dutch nominal infinitives Like many languages Dutch has expressions showing a mix of nominal and verbal characteristics: the nominal infinitives. In Dutch they come in three types, reflected in the choice of determiner: 1. Het-determiner (the neuter form of the definite article) 2. Dat-determiner (the neuter form of the demonstrative) 3. No determiner Schoorlemmer (2001) discusses types 1. and 2., illustrated in (1a,b). (1c) illustrates type 3. (1)
a.
Het oproepen
van getuigen door de officier
the summon-INF of b.
the coroner
Dat afschuwelijke overlast veroorzaken van jou that terrible
c.
witnesses by
trouble cause-INF
of you
Bomen kappen (door de industrie) is schadelijk trees
felling (by the industry)
is harmful
Schoorlemmer notes a number of interesting contrasts between type 1 and type 2 infinitives, which I will not discuss here, and further presents an overview of nominal and verbal properties of nominal infinitives, as identified in the literature (Van Haaften et al., 1985; Hoekstra and Wehrmann, 1985; Reuland, 1988; Hoekstra, 1997). Nominal infinitives share with canonical DPs the following properties:
They are introduced by a determiner. They can be modified by adjectives. A subject or object may be expressed in a van-phrase (NB: there can be only one van-phrase). The van-phrase always occurs in a post-head position. The following properties are shared with VPs:
They can be modified by adverbs. Argument PPs can precede or follow the head. The subject can be realized as a door-phrase (by-phrase). (Bare accusative) direct objects must precede the verb. They may contain complex verbal structures. A number of properties deserve specific attention:
The direct object can be realized both preverbally as an NP or postverbally as a van-PP (of-PP). Modifiers can be realized with a schwa marking them as adjectives, or without a schwa marking them as adverbs. One and the same construction can be modified both by an adjective and an adverb, but only in that order. The presence of a determiner is necessary to license modification by an adjective (marked with a schwa). This entails that adjectives are impossible in type 3 nominalizations as illustrated in (2): (2)
a.
*FrequenteADJ/frequentADV bomen kappen (door de industrie) is schadelijk. frequent/frequently trees felling (by the industry) is harmful
b.
*FrequenteADJ/frequentADV kappen van bomen (door de industrie) is schadelijk. frequent/frequently felling of trees (by the industry) is harmful
Some further examples of type1 showing the main pattern are given in (3) and (4): (3)
a.
Het frequenteADJ/frequentADV bomen kappen (door de industrie) is schadelijk.
b.
*Het frequenteADJ/frequentADV kappen bomen (door de industrie) is schadelijk.
c.
Het frequenteADJ/frequentADV kappen (van bomen) (door de industrie) . . .
The frequent/frequently trees felling (by the industry) is harmful The frequent/frequently felling trees (by the industry) is harmful The frequent/frequently felling of trees (by the industry) is harmful
E. Reuland / Lingua 121 (2011) 1283–1296
d.
*Het frequenteADJ/frequentADV van bomen kappen (door de industrie) . . .
e.
Het *frequenteADJ/OKfrequentADV door de industrie bomen kappen is schadelijk.
1285
The frequent/frequently of trees felling (by the industry) is harmful The frequent/frequently by the industry trees felling is harmful f.
*Het frequenteADJ/frequentADV door de industrie kappen van bomen is schadelijk.
g.
Hun frequenteADJ/(?)frequentADV bomen kappen is schadelijk
The frequent/frequently by the industry felling of trees is harmful Their frequent/frequently trees felling is harmful h.
Hun frequenteADJ/frequentADV kappen van bomen is schadelijk Their frequent/frequently felling of trees is harmful
(4)
Hun frequenteADJ/*frequentADV kappen (door/*van de industrie) is schadelijk Their frequent/frequently felling (by the industry) is harmful (POSS is OBJ)
However: How verbal is verbal, how nominal is nominal? First of all, consider the question of how verbal is verbal in the light of the puzzle posed by Argument optionality. ‘Truly’ verbal and nominal complementation show the following contrast: A suitable discourse can license absence of the ‘obligatory’ complement in nominals, but not with true verbs. (5).
a.
De bomen moesten allemaal weg. Het vellen kostte veel tijd. The trees all had to go. The felling cost a lot of time.
b.
De bomen moesten allemaal weg.* Jan velde gedurende drie uur The trees all had to go. John fell for three hours.
(6)
a.
Dit struikgewas is schadelijk. Frequent verwijderen (door een expert) is nodig. This shrubbery is harmful. Frequent removing (by an expert) is needed.
b.
Dit struikgewas is schadelijk. *Een expert moet frequent verwijderen. This shrubbery is harmful. An expert should frequently remove.
(7)
a.
De stad viel in de handen van de vijand. Het vernietigen/de vernietiging kostte veel tijd. The town fell into the hands of the enemy. The destruction took a lot of time.
b.
De stad lag op de route van het vijandelijke leger. De vernietiging door de vijand kostte echter veel tijd./Het vernietigen door de vijand kostte echter veel tijd.
c.
De stad viel in de handen van de vijand.*De vijand was lang bezig te vernietigen.
a.
Emoties zijn zwaar. Frequent uitdrukken is belangrijk. Ongeremde uitdrukking is goed.
The town was in the way of the enemy army. The destruction by the enemy took a lot of time, however. The town fell into the hands of the enemy. The enemy took a long time destroying. (8)
Emotions are heavy. Frequent expression is important. Uninhibited expression is good. b.
Emoties zijn zwaar. *Jan drukte ongeremd uit omdat dat belangrijk was.
a.
Deze problemen zijn onoplosbaar. Veelvuldig opgeven moet vermeden worden.
Emotions are heavy. John uninhibitedly expressed because it was important. (9)
These problems are unsolvable. Frequent assignment should be avoided. b.
Deze problemen zijn onoplosbaar. *We geven desondanks constant op. These problems are unsolvable. Nevertheless we assign constantly.
These examples show that Dutch has considerable freedom in the realization of arguments in nominal infinitives, and that arguments that are syntactically obligatory in verbal constructions are not syntactically obligatory in the corresponding nominal infinitives. This is independent of the character of the modifier, whether or not a ‘subject’ is realized as a POSS phrase or a by-phrase, etc. This property obtains throughout in all other types of nominalization. The question is why this is so. A possible answer would be that these constructions are fully nominal and that nouns do not have obligatory arguments (contra Grimshaw, 1990, but see below for discussion of Grimshaw’s facts). However, this leads to the following further questions: (i) What does it mean to be nominal? (ii) Why would this property hold of nominals? These are the main questions I will start addressing now.
E. Reuland / Lingua 121 (2011) 1283–1296
1286
3. What does it mean to be a nominal? The question of what is means to be nominal is a central issue in the theory of syntactic categories. Giving an overview of the discussion of syntactic categories in the literature up to the present would lead me beyond the scope of this article (see Vinokurova, 2005 for such an overview and illuminating discussion). I will take as my starting point certain issues as they arise from the current framework of Distributive Morphology (Halle and Marantz, 1993; Marantz, 1997; Embick, 1997; Harley and Noyer, 1998; Harley and Noyer, 1999; Borer, 2003, etc.). This framework is based on the following tenets1: (10)
Lexical roots are category neutral, they are assigned a category X by merging with a small x. n+H=N a+H=A v+H=V
However, this tenet does not express what it means to be an n, rather than a v or a. It effectively states the following: - any root can merge with nominalizer to become a noun, a verbalizer to become a verb, an adjectivalizer to become an adjective. - there are no asymmetries between members or lexical categories. - all lexical categories are morphosyntactically complex. Vinokurova (2005) argues that in its generality, this view cannot be maintained. In particular, she shows it does not hold for Sakha, a Turkic—agglutinating—language spoken in the Sakha republic (Russia, East Siberia). For languages such as English with extensive 0-morphology it only apparently applies. Instead, she argues that a distinction exists between primitive verbs and primitive nouns. Consider cases of Noun–Verb pairs as in (11): (11)
a carpet – to carpet, an urchin – to urchin, a cut – to cut, a break – to break
In languages such as Sakha there is a direction of derivation as given in (12): (12)
a.
(N ! V) a carpet ! to carpet, an urchin ! to urchin;
b.
(V ! N) to cut ! a cut, to break ! a break.
In Sakha nouns such as those in (12a) and verbs as in (12b) are simplex. The corresponding verbs in (12a) and nouns in (12b) are morphologically complex involving ‘carpetnoun’, ‘urchinnoun’, ‘cutverb’, ‘breakverb’ plus some affixes. This is illustrated in (13): (13)
Noun member: structure
Verb member: structure
Vinokurova
N
N-Verbalizer
DM/XS
H-nominalizer
H-verbalizer
Example
Uot ‘fire’
Uot-taa ‘fire’
from
Bas ‘head, master’
Bah-yj ‘master’
Sakha
Baqa ‘desire’
Baqa-r ‘desire’
Vinokurova
V-Nominalizer
V
Example
Tu¨m-u¨k ‘conclusion’
Tu¨m ‘conclude’
from
Kepsee-n ‘story, narration’
Kepsee ‘tell, narrate’
Sakha
Taraa-x ‘comb’
Taraa ‘comb’
1 See for instance the following summary in Harley and Noyer (1999): Specifically, the different ‘parts of speech’ can be defined as a single l-morpheme, or Root (. . . .) in certain local relations with category defining f-morphemes. For example, a ‘noun’ or a ‘nominalization’ is a Root whose nearest c-commanding f-morpheme (or licenser) is a Determiner, a ‘verb’ is a Root whose nearest c-commanding f-morphemes are v, Aspect and Tense (. . . .). In more recent work, Borer gives a detailed analysis of the role of event structure in the mapping procedure, which I cannot possibly do justice to here. In essence her approach stays within the limits of what I sketch here.
E. Reuland / Lingua 121 (2011) 1283–1296
1287
The reverse derivations are simply impossible. What is needed is an independent characterization of what is nominal and verbal to explain the direction of derivation. An interesting proposal for such an independent characterization is given in Baker (2003). Baker proposes the following fundamental characteristic for nominals: Nominals satisfy Geach’s criterion of identity They can be used in same-contexts: this tree is the same as that one, etc. Syntactically this property is encoded as being carrier of an index. Baker acknowledges that indices violate inclusiveness. Therefore, he suggests that indices are added at the CI interface. This leads to the question of how syntax knows that a noun is a noun (again, independently from some category label, since it is the latter we wish to reduce). Some item x cannot be classified as a noun or a non-noun on the basis of inspection alone. Note, furthermore, that the criterion of identity is ill-suited to determine what is nominal in nominalizations, and certainly dubious in eventive expressions, as illustrated in (14): (14)
a.
? this walk is the same as that one
b.
*The felling of the trees by John is the same as the felling of the trees by Bill
If so, this leads us once more to the question of what it is that is nominal in nominalizations. (See Baker, 2003 for an overview of problems in using other criteria, e.g. morphological criteria like types of affixation.) What we need is a criterion whose applicability can be determined by inspection. Vinokurova (2005) presents the following important insight: - verbs are relational, nouns are not relational, adjectives are properties Since prima facie not all verbs are relational (witness passives and unaccusatives), it is necessary to make this notion precise. To do so, I will discuss the implementation of this idea within the theory of thematic roles presented in the Thetasystem developed by Reinhart (2002), Reinhart and Siloni (2005), Marelj (2004), and others. As we will see, one of the core ideas of that theory is the active lexicon, wherein there are lexical operations that may derive unary predicates from underlyingly relational concepts. 4. The Theta-system: the central system of concepts I will start with a concise overview of the main properties of the Theta-system. Its basic tenets are the following: (i) Abstract concepts need to be formally coded in order to be legible to the Computational System (CS); (ii) The Theta-system enables the interface between the system of concepts and the CS as well as the interface between the system of concepts and the inference system. The Theta-system is the narrow-channel by which the computational system can access information from the conceptual system. Whatever conceptual information cannot pass through this channel cannot be input to the syntactic computation. The Theta-system comprises the components in (16): (16)
Components of the Theta-system a.
A system for formally encoding the information about abstract concepts the computational system must be able to access: the linguistic coding of causal relationships;
b.
Lexical entries (i.e. particular coded concepts);
c.
Operations on entries;
d.
Merging instructions (linking/mapping rules).
I will briefly sketch these in turn. 4.1. The system The conceptual information the computational system can see is in the form of theta-roles. All theta roles are formally coded as clusters defined in terms of two binary features: [c] (cause change), [m] (mental state).
E. Reuland / Lingua 121 (2011) 1283–1296
1288
A cluster need not be specified for (all) features. This allows the null-cluster as a limiting case.2 The features are characterized in (17): (17)
The features a.
The c-feature The feature/+c represents that the role is perceived as a sufficient condition (Shen, 1985). The feature/c represents that the argument is identified as not being the cause. If the c-feature is not represented, the causal status of the argument is undetermined.
b.
The m-feature The/+m feature is related to the role of the mental state of the participant present in the relation ‘motivate’. Hence, the/+m feature entails animacy. A/m specification does not entail in-animacy.
4.2. Lexical entries The thematic properties of Verbs (i.e. n-place predicates) are encoded in terms of these feature clusters. The following definitions apply: (18)
Definitions and notations [α]
= Feature cluster α.
/α
= Feature (and value) α. (E.g. the feature/+m occurs in the clusters [+c + m], [-c + m], and [+m])
[/α]
= A cluster one of whose features is/α. (E.g. [/-c] clusters are [-c + m], [-c-m] and [-c].)
[+]
= A cluster all of whose features have the value +.
[]
= A cluster all of whose features have the value -.
These definitions are necessary to formulate well-formedness conditions on lexical entries, such as the Cluster Distinctness Constraint (CDC), which states that two indistinct theta-clusters cannot be both realized on the same predicate. (The CDC rules out realizations of a verb like anger as in The article angered Bill at the government, see Reinhart (2002) for more discussion). Importantly for our present purposes, they are needed to specify the lexical operations discussed in section 3.3, and the linking conditions to be discussed in section 3.4. The Theta-system specifies a restricted number of operations on lexical entries expressing significant relations between lexical items. This part of the theory will be sketched next. 4.3. Operations on lexical entries The lexical operations in the Theta-system are tied to the following principle: (19)
Lexicon Uniformity Hypothesis Each verb-concept corresponds to one lexical entry with one thematic structure. The various thematic forms of a given verb are derived by lexicon operations from one thematic structure.
This principle entails that the unaccusative forms in (20b) and (21b) are derived from the basic transitive forms in (20a) and (21a). (20)
2
a.
Peter melted the ice.
b.
The ice melted.
The two binary features define nine feature clusters:
(1)
[+c+m]
Agent
(2)
[+c-m]
Instrument
(3)
[-c+m]
Experiencer
(4)
[-c-m]
Theme
(5)
[+c]
Cause
(6)
[-c]
Recipient Goal/Benefactor
(7)
[-m]
Subject Matter/Source
(8)
[+m]
Sentient
(9)
[-]
Arb in middles
E. Reuland / Lingua 121 (2011) 1283–1296
(21)
a.
Peter froze the ice.
b.
The ice froze
1289
Reinhart assumes that unaccusatives are lexically derived also when the basic transitive form is not realized in a language.3 The available operations on verbal concepts are: i. Reduction ii. Saturation (e.g. Passive) iii. Entry-changing (e.g. Lexical Causativization) To give the reader some background about the way the system works, I summarize its main operations. This is important to enable the reader to assess what is involved in the nominal-verbal contrast. i. Reduction Reduction affects the number of positions in the grid of a lexical element that are linked to an argument bearing a thematic role. The way it operates depends on whether an internal or the external role is affected. (i) Reflexive (unergative) entries are the output of reduction of the internal role. (ii) Unaccusative entries are the output of reduction of the external role. The two are not entirely symmetrical. When the internal role is reduced, it retains its semantic effect. The result is a composite role rather than just the original external role, as illustrated in (22). (22)
Internal Reduction/Bundling: a. Vacc (θ1, θ2) ! Rs(V) (θ1,2) b. Rs (V)(θ1,2) $ θ1,2 (λx (V (x,x))) c. shaveacc([+c + m]1,[-c-m]2: Lucie shaved him. d. Rs(shave)([+c + m]1): Max shaved.
When the external role is reduced, it is no longer semantically visible, as we see in (23): (23)
Expletivization: Reduction of an external [+c] role (semantically null function) a. Vacc (θ 1[+c], θ2) ! Re V(θ2) b. Re(V) (θ2) = V(θ2) c. openacc ([+c], [-c-m] ! Re(open)[-c-m] d. worryacc ([+c], [-c + m] ! Re(worry)[-c + m]
The following conditions on the availability of these operations should be noted: (i) Reduction can apply only in a two-(or more)-place entry, one of whose roles is external (a [+] cluster). (ii) Reduction eliminates the accusative feature of the verb (fully or partially, see for discussion below). (iii) Reduction and saturation (passive) are exclusive (only one can apply at each entry). Consider next the operation of saturation. ii. Saturation A crucial component of saturation is variable binding: A variable in the representation of the conceptual structure is bound by some operator, e.g. an existential or generic operator (E or Gen). Saturation entails that the theta-role is unavailable for syntactic purposes. An example is passivization. Passive saturation closes existentially the external argument of a verb as illustrated in (24): (24)
a. b.
wash (θ1, θ2) Ex (wash (x, θ2)) Max was washed
! Ex (x washed Max)
In addition to reduction and saturation, there are also entry-changing operations. iii. Entry-Changing Operations An example is lexical causativization, illustrated in (25) and (26): (25) (26)
a. b.
The dog walked ! She walked the dog. The horse jumped ! The rider jumped the horse over the fence
The components of this operation are given in (27). It involves both changing the content of a cluster and adding a cluster: 3
I will come back to this below.
E. Reuland / Lingua 121 (2011) 1283–1296
1290
(27)
Lexical Causativization a.
Decausativize: Change a/ + c feature to a/-c feature.
b.
Agentivize: Add an agent role.
walk: ([+c + m]) ! walk: ([-c + m]) walk: ([-c + m]) ! walk: ([+c + m], [-c + m]) Languages may differ as to where such operations take place. This variation is expressed in the Lex-Syn Parameter: UG allows thematic arity operations to apply in the lexicon or in the syntax. Concept forming operations can only take place in the lexicon. Some further variation takes place in the instructions for merging and linking arguments to the theta-roles projected by the lexical element. 4.4. Merging/linking instructions Linking instructions are stated in terms of s-selection (i.e. the type of arguments that are encoded on the verb’s grid) and the following set of rules and generalizations: (28)
Lexicon marking Given an n-place verb-entry, n > 1, a.
(29)
(30)
Mark a [] cluster with index 2.4
b.
Mark a [+] cluster with index 1.
c.
If the entry includes both a [+] cluster and a fully specified cluster [/α,/-c], mark the verb with the ACC feature.
Relevant generalizations of lexical operations: a.
Saturation and reduction apply to the marked entry (i.e. after marking).
b.
Reduction eliminates the accusative feature of the verb (fully or partially).
CS merging instructions. a.
When nothing rules this out, merge externally.
b.
An argument realizing a cluster marked 2 merges internally; an argument with a cluster marked 1 merges externally.
The way reduction affects Case reflects another source of cross-linguistic variation. As we saw in (22), repeated here, in English the ACC feature is fully eliminated by reflexive reduction: (22)
Internal Reduction/Bundling: a.
Vacc (θ1, θ2) ! Rs(V) (θ1,2)
b.
Rs (V)(θ1,2) $ θ1,2 (λx (V (x,x)))
c.
shaveacc([+c + m]1,[-c-m]2: Lucie shaved him.
d.
Rs(shave)([+c + m]1): Max shaved.
The object position of shave in (22d) is just empty. Dutch and many other languages do not allow this position to be empty. Rather, a simplex anaphor is required, as in the Dutch equivalent of (22d) Max scheert zich. Here, the simplex anaphor zich must be inserted to check a residual structural Case. Another way of saying this is that in Dutch, the reflexive entry of scheren is still syntactically 2-place. The internal role reduction only eliminates part of the Case frame, the thematic part is eliminated, but the structural Case component is still there and requires checking.5 For sake of concreteness, I give a sample derivation of a transitive structure in (31) and an unaccusative derivation in (32). (31)
Transitive Derivations a.
Max ate an apple.
b.
Base entry: eat ([+c + m], [-c-m])
4 Note that these indices are just annotations for the merging instructions to apply. They should not be confused with referential indices, which the inclusiveness condition prohibits. 5 This presupposes a componential analysis of ACC case, into structural and thematic. See Belletti (1988) for an earlier version of this idea. For extensive discussion from the current perspective, see Reuland (2011).
E. Reuland / Lingua 121 (2011) 1283–1296
(32)
c.
Marking: eatacc ([+c + m]1, [-c-m]2)
d.
Merging: By (30b), ([+c + m]1) merges externally, ([-c-m]2 merges internally
1291
Unaccusative Derivations a.
The vase broke.
b.
Base entry: break ([+c], [-c-m])
c.
Marking: breakacc ([+c]1, [-c-m]2)
d.
Expletivization applies to the marked entry, giving rise to (break ([-c-m]2))
e.
Merging: Internal, by (30b)
Just as in the case of internal reduction/reflexivization, languages may vary as to whether ACC is eliminated by the reduction operation, or whether a Case residue remains. If a case residue remains, it has to be checked. Checking can take place by a simplex reflexive element, like Dutch zich, by a clitic, or some other marker. It is not always straightforward to determine why precisely a marker has to be present, and what it contributes. In German we see the element sich with middles, reflexives and unaccusatives. In Italian we see the clitic si again with middles, reflexives and unaccusatives, but also with passives. The same properties are found with the reflexive clitic se in SerboCroation. The question is then what unifies such seemingly disparate uses. What these cases share is that an operation on argument structure has applied that is licensed by sich/si/se. In Sakha we see a similar pattern. As discussed extensively in Vinokurova (2005), Sakha has a range of reflexive, passive and also a yet wider range of unaccusative entries that require the presence of a special affix –n. For sake of illustration, (33) gives some causative/unaccusative pairs from Vinokurova (2005, App. IV. II): (33)
Causative
Causative Gloss
Unaccusative
Tiree
Prop, support, lean
Tiren
Aqaa
Smear, glue
Aqan
Tox
Spill
Toqun
From a synchronic perspective, Sakha is best described as a language lacking structural Case. Vinokurova concludes that the most straightforward generalization is that –n in Sakha marks the application of a (reduction or saturation) operation on argument structure. How, precisely, it contributes to licensing would be an interesting topic for further research.6 For current purposes it suffices that Sakha clearly shows morpho-syntactic effects of reduction in a wide range of unaccusative entries. 5. Elaboration 5.1. 1-place verbs Coming back to our main concerns, as we saw, the Theta-system derives unaccusative entries from 2-place entries. Vinokurova (2005) now proposes to generalize this property. That is, what sets apart basic verbs from basic nouns is that all true verbs have a 2-place source, and basic nouns do not. This tallies with the analysis of unergative verbs, the other type of prima facie 1-place verbal predicates, put forward in Hale and Keyser (1993, 1998, 2000). These authors analyze unergative verbs as denominal: transitive structures involving a verbalizer and a nominal complement. Following this line, Vinokurova proposes that yet another class, namely weather-verbs, also be analyzed as unergatives.7 This leads to what I will refer to as Vinokurova’s thesis: 6 The proposal that unaccusatives are systematically derived from two-place entries by a reduction operation whose effects show up in particular markings is orthogonal to the morpho-syntactic manifestation of transitivity. Leon Stassen (personal communication) mentions that in some languages transitives appear to be derived from 1-place verbs in the sense that they are morphologically more complex. As always, languages do not bear their analysis on their sleeves. What would be needed for firm conclusions is an in-depth investigation of the nature and distribution of such morphemes. Whatever complexity shows up could well be the reflection of a Case frame, or object agreement. In principle, languages could exist in which a substantial class of ‘ordinary’ transitive concepts have obligatory external argument reduction, with only unaccusatives surfacing, an inventory that would then be supplemented by the operation of an entry-changing causativizing affix as discussed above. Mutaka and Safir (2008) (M&S) describe a very intriguing phenomenon in Kinande. Kinande, -nab- ‘‘wash’’ and -hakab- ‘‘smear’’ have intransitive reflexive interpretations like their counterparts in many other languages. As in English there is a corresponding transitive reflexive form. In English, the transitive reflexive is formed by using a self-form (himself, etc.). In Kinande the productive strategy is not available, unless the verb stem is extended with the morpheme –i. The question is how this fact is to be interpreted. M&S argue that this indicates a direction of derivation opposite to what Reinhart and Siloni claim for English. M&S discuss a number of interesting further properties of this morpheme, without reaching a definitive conclusion about its properties. As they say, their approach to -i- is somewhere between those that treat its Bantu cognates simply as causative or direct causative (. . . ) and those that treat it as strictly a transitivizer. Unlike what they suggest, I do not think this result is necessarily a problem for Reinhart & Siloni’s approach. If the i-affix is a causative verb extension, such a fact is compatible, as we saw. Moreover, we know independently that many languages do mark unaccusatives. Thus, taking all the facts together the fundamental question appears to be what would motivate a see-saw effect in Kinande and similar languages. Such an investigation would lead us beyond the scope of the present contribution, however. 7 Vinokurova does not discuss the status of theme-unergatives, which include emission verbs such as glow. It would seem that they are amenable to the same analysis.
1292
(34)
E. Reuland / Lingua 121 (2011) 1283–1296
Vinokurova’s thesis: Verbs represent intrinsically relational concepts in opposition to verbs: noun concepts are intrinsically non-relational Adjectives are intrinsically 1-place concepts
5.2. (Relational) nouns If noun concepts are intrinsically non-relational, what about nouns such as sister and father? These are generally analyzed as relational: a sister is necessarily somebody’s sister, a father somebody’s father. Vinokurova argues that the relevant notion of being relational is made explicit by the properties of theta-clusters: A verbal concept like EAT is associated with two necessary conditions – someone to perform the eating process (also a sufficient condition) and something to be eaten. These causal relations are translated into θ-features yielding two arguments of the verb eat. With relational nouns like sister, on the contrary, it is not motivated in terms of causality that they should take an argument. This is because a necessary and sufficient condition for being a sister is being a female sibling. However, the concepts FEMALE SIBLING and SISTER are coextensive. Therefore, with relational nouns it is impossible to translate their causal relations into distinct feature clusters – the potential arguments. (Vinokurova, 2005:77). Hence, basic nouns do not have a theta-grid in the relevant sense. This entails that they cannot undergo lexical operations defined on theta-grids. 6. How to proceed from here? This being said, we still have to address our main problem: What is nominal in nominalizations? I will briefly discuss three options. Option 1: The status of an element as nominal or verbal can be determined by inspection of the concept. This entails that the categories V, N and A become predictable, hence redundant. Functional categories select for a lexical item with a relational conceptual structure (!V), a non-relational conceptual structure (! N), or a 1-place conceptual structure (! A). Since the arity reducing operation of Reduction/Bundling applies to the marked concept, derived 1-place verb-concepts reflect that they are derived from a relational verb/concept. However, if this is all, the status of nominalizations, which somehow combine nominal and verbal characteristics, is still left open. Let’s therefore consider a further option. Option 2: Reconsider one part of the procedure in the theta-system. According to the original version of the theta-system, the argument structure of verbal concepts is made readable for CHL by channelling it through the [c], [m] system. It assumes that being verbal is given. Consider now the following addition: Being channelled through the feature system is what makes a concept verbal. The [ c], [m] system articulates the concept. If a concept is not relational, channelling it through the [c], [m] system can either be assumed to apply vacuously, or be considered as undefined. In any case the concept does not become relational. If a concept is relational it can, but need not be articulated by the [c], [m] system. If yes, it is realized as verbal. If not, it is realized as nominal. If so, categorization as V, N and A is again predictable. It can be read off, and need not be annotated. Consider, for instance, the concept vellen ‘fell’. It is potentially relational. If channelled though the [c], [m] system, its theta-clusters will be articulated, and it will be realized as a verb. Having articulated theta-clusters is visible by inspection, and its theta-clusters will be subject to the merging procedure. If not channelled through the [c], [m] system, it has no clusters that are subject to the merging procedure; hence, its arguments are realized optionally. A potential problem with this line is that one would expect that ‘arguments’ of nouns/nominalizations may connect (in some way or other) with any meaning aspect of the concept. Arguments of nouns would be able to bear ‘roles’ that cannot be encoded in the [c], [m] system, hence, are unlike any theta-roles of verbal arguments. There is no evidence that this obtains. Let us therefore consider a yet more restrictive option: Option 3: The essential difference between verbs and nouns resides in the availability in principle of merging instructions. The roles that nouns can assign (if they have the right conceptual make-up) are just the roles that verbs can assign; however, they are not accessible to marking in the sense of (28). This entails that (30) is undefined and leads the following modification of Vinokurova’s thesis:
E. Reuland / Lingua 121 (2011) 1283–1296
(35)
1293
Vinokurova’s thesis modified:
Verbs represent relational concepts for which merging instructions are defined
Nouns represent concepts for which merging instructions are not defined
(Suggestion: Adjectives represent concepts with merging instructions for one argument)
In this form it is applicable both to basic nouns and to ‘nominalized’ verbs. Selection still holds in relation to nominals, which entails that arguments can be realized using appropriate prepositions, and is consistent with Grimshaw’s (1990) thesis that nouns are deficient theta-markers. (35) has the following consequence: If no merging instructions for the complements of a lexical item are defined, its complements cannot be obligatory. As an anonymous reviewer notes, nouns do mark their arguments, for instance, the direct object by the preposition of in the destruction of Rome, or by the genitive as in Rome’s destruction, or the Vandal’s destruction of Rome. Thus, one may wonder whether nouns would not have merging instructions after all. Interestingly, Higginbotham (1983) shows convincingly that the element in the POSS position does not bear a theta role determined by the noun in the same way as a subject bears a theta-role determined by the verb. Rather he argues, the role it receives in interpretation is determined by a general aboutness-relation, as for instance illustrated by the use of expressions like yesterday in yesterday’s examination. Just like no merging instruction of the type specified in (28–30) says that yesterday has to be ‘‘merged externally’’ to realize a particular theta cluster, no merging instruction specifies that the Vandals would have to merge externally in order to realize a particular cluster in The Vandal’s destruction of Rome, and in fact, it can equally well express agenthood in a by-phrase as in The destruction of Rome by the Vandals. Thus, clearly, the concept destroy allows for an agent. An agent relation can be picked up in the POSS position, by virtue of Higginbotham’s aboutness strategy, or in a position where this strategy is not available on the basis of a preposition marking the role. The same applies to the theme. Conceptually, it is available, but it is not associated with a particular merging instruction. For an argument in POSS position it is available by aboutness; for another position it must be explicitly marked by the preposition of. As noted above, this is in line with Grimshaw’s deficiency thesis. We now have an independent characterization of what’s nominal in Dutch nominal infinitives. Do we also have an explanation for the optionality of their arguments? I am grateful to an anonymous reviewer for bringing this question up. Clearly, there are different levels of explanation. Let us see how far we have come. Blindly applying the Theta-system we can say that it has been explicitly developed as a theory about verbal concepts, crucially relying on an articulation in terms of the features [c], [m]. We have an independent characterization of the representation of nominal structures as not being articulated in terms of these two features. As they are formulated, there is no way the merging instructions (28–30) could apply to such nominal structures. There is every reason to think that they do not, so we have an explanation at the level of a particular theory. Of course, it is always extremely useful to move beyond this and ask ourselves deeper why-questions. In most cases, and to my knowledge in this case as well, this requires moving beyond what we have explicit theories of, hence coming closer to speculation. But let’s give it a try. The core idea behind Reinhart’s feature system is that causation and mental involvement are basic categories in which we organize the world. Note the term organize. This organization is not inherent in reality, but in the way we may choose to represent reality, but are not obliged to. Thus, a destruction or examination or felling event can be represented in different ways. However, once we choose to do so in terms of causation and mental involvement the representation becomes necessarily relational, either in terms of our relation to the eventuality (mental involvement) or in terms of the internal organization of the eventuality (causation). This leads us back to Vinokurova’s core intuition: the relational character of verbs, versus the non-relational character of nouns. Of course, there are further questions to ask. Why do verbs – abstracting away from big V or little v’s – check structural accusative, and why is it that nouns do not? Or to put it differently, why are structural Cases (nominative/accusative) in some broad sense related to the Tens/aspect system? What is the relation between the dependent character of nominativeaccusative (see Marantz, 2000) and the eventive/relation character of verbs? These are all important questions, but any attempt at discussion would carry us beyond the scope of this contribution. Let us go back to Vinokurova’s thesis as we revised it in (35), and the way it accounts for the Dutch facts we started out with. In this form, it contradicts Grimshaw’s (1990) analysis of complex event nominals. That’s why the last section is devoted to this issue.
7. Complex event nominals in English8 Grimshaw (1990) argues that complex event nominals have argument structure, just like verbs. She claims that complements to complex event nominals are obligatory as in the case of the corresponding verb. The view that complements 8
I would like to thank Anna Asbury and Bill Philip for their help.
1294
E. Reuland / Lingua 121 (2011) 1283–1296
in nominalizations are not obligatory is due to the fact that many derived nominals are ambiguous between the complex event reading and a simple event/result reading. In her view, Given a proper disambiguation, the obligatoriness can be observed. The main difference between complex event nominals and verbs: Nominals are deficient theta-role assigners and require help from a preposition. Grimshaw presents the following examples of unambiguously complex event nominals and the obligatoriness of their arguments: (36)
a.
The felling *(of the trees)
b.
They felled *(trees)
c.
The destroying *(of the city)
d.
They destroyed *(the city)
Disambiguation by a modifier is shown in the following examples: (37)
(38)
a.
The expression is desirable
b.
*The frequent expression is desirable
c.
The frequent expression of one’s feelings is desirable
d.
We express *(our feelings)
a.
The assignment is to be avoided
b.
*The constant assignment is to be avoided
c.
The constant assignment of unsolvable problems is to be avoided
d.
We constantly assign *(unsolvable problems)
The effect of disambiguation by agents is shown in (39) and (40): (39)
a.
*The instructor’s intentional/deliberate examination took a long time.
b.
The instructor’s intentional/deliberate examination of the papers took a long time.
(40)
a.
The destruction was awful to see
b.
*The enemy’s destruction was awful to watch
c.
The enemy’s destruction of the city was awful to watch
d.
The enemy destroyed *(the city)
These facts appear to be unexpected given the Dutch facts from section 1. This leads to the question of how different English is from Dutch. However, properly considered, English complex event nominals are not very different at all. In fact, according to my informants, they behave just like Dutch. In (36)’–(40)’ I use English examples taken from the list above in the same context I used in the Dutch cases. The effect is striking: (36)’ (37)’ (38)’ (39)’ (40)’
a.
There were a lot of trees on the building lot. The felling took several days.
b.
There were a lot of trees on the building lot. *The workers felled for several days
a.
Suppressed emotions are dangerous. Frequent expression is desirable.
b.
Suppressed emotions are dangerous. *Therefore we expressed frequently
a.
These problems are unsolvable for these students. Regular assignment is to be avoided
b.
These problems are unsolvable for these students. *Therefore, we don’t assign regularly.
a.
The papers were extremely tough reading. The instructor’s careful examination took a long time.
b.
The papers were extremely tough reading. *The instructor carefully examined for a long time.
a.
The city was in the way of a strong enemy army. Yet, the complete destruction by the enemy took several days.
b.
The city was in the way of a strong enemy army. *The enemy completely destroyed, which took them several days.
A clear contrast emerges between arguments of complex event nominals and arguments of verbs: given an appropriate discourse, arguments that are obligatory for verbs can be omitted in nominalizations. Hence, English complex event
E. Reuland / Lingua 121 (2011) 1283–1296
1295
nominals are as compatible with Vinokurova’s modified thesis as their Dutch counterparts. This leads us to the question of what types of variation between nominals one may expect, as expressed in the title of the next section.9 8. Expected types of variation between nominal(ization)s I will be content giving just a list of types, with their characteristics. Although, no doubt, each would merit a more extensive discussion than I can give here, for present purposes this overview will suffice: 1. Basic nouns No internal relational structure ! no merging instructions ! nominal 2. Simple event nominals An affix selects a relational concept without articulated theta-structure ! no merging instructions ! nominal 3. Complex event nominals An affix selects a relational concept with articulated theta-structure ! blocks assignment of merging instructions ! nominal 4. Nominal infinitives An affix selects a relational concept with articulated theta-structure and (thematic) ACC case, and blocks assignment of merging instructions ! nominal 5. Gerunds A ‘‘nominalizing affix’’ that applies in the syntax leaving a full verb with merging instructions in its domain ! it is inflectional rather than derivational (gerundival –ing, etc.) ! not nominal Instantiates a class of elements licensing verbal projections to appear as arguments, functionally similar to complementizers. This overview leads to the following further questions: (i)
(ii)
a.
Do we have nominalization as a lexical operation on items that are fully verbal in the sense that they have explicit merging instructions for their arguments and that ‘takes away’ the merging instructions?
b.
Do we have nominalization as a syntactic operation on items that are fully verbal in the sense that they have explicit merging instructions for their arguments and that ‘takes away’ the merging instructions?
How precisely do nominalizers affect the item they merge with? I see the following options:
(i)
An affix may suppress (using Grimshaw’s term) an argument. This would entail that the concept becomes nonrelational and that merging instructions are not applicable. This requires that suppression applies, before lexicon marking (28) can apply. However, this can only account for some of the above types.
(ii)
Alternatively, the affix projects a structure to which the merging instructions cannot be applied. Viability of this option depends on implementation of the merging instructions.
Both require further research for a proper assessment. However, these relatively open questions notwithstanding, I feel we have come a long way towards an understanding of the question we started out with, which leads me to a brief final section. 9. Conclusion I would like to conclude saying that Vinokurova’s thesis modified explains the following: What is nominal in nominalizations Why arguments are optional in nominalizations Which is what we set out to do. 9 An anonymous reviewer notes that the data are "not as cut and dried" as they are presented here. The reviewer acknowledges that there is a contrast in the direction I found, but in many cases (s)he finds the sentences with argument omission more questionable than my informants. So, apparently there is idiolectal variation here that merits further investigation. Another interesting fact the reviewer brings up is that " –ing-nominals lacking overt arguments are virtually as bad as sentences even with contextual information provided", as in (i):
(i)
Repressed emotions are generally considered to be unhealthy. Most experts therefore recommend frequent expressing *(of them/of such emotions)
The interesting thing is that here the reviewer sees a clear contrast between the sentence with the –ing form expressing and the one with expression. It is well-known that certain English –ing-forms are fully verbal (Reuland, 1983). The reviewer’s observation suggests that for some English speakers the –ing forms are predominantly construed as containing a verbal projecting (with a PRO subject) also when embedded in a functional shell with DP properties licensing the adjectival form of the modifier and the of-form of the objective Case (except for (36a’) which the reviewer finds fully acceptable), that is as gerunds, which are listed as verbal in section 7.5 rather than as nominalizations. Clearly, there is a form of competition between construing an –ing-form as a gerund or as a nominalization if their further properties leave this open. What factor determines whether an –ing-form is construed as one or the other is again a matter for further investigation.
1296
E. Reuland / Lingua 121 (2011) 1283–1296
Acknowledgements Many thanks to Jaklin Kornfilt and John Whitman for organizing a very inspiring workshop and inviting me to participate and present my ideas. I am also indebted to the audience for their great talks, and stimulating discussion. Last but not least, I am very grateful to two anonymous reviewers for their very useful comments that helped me improve the final article both in form and content, and again to Jaklin and John for all their work in getting this collection published. Needless to say that I am solely responsible for all errors. References Baker, M., 2003. Lexical Categories: Verbs, Nouns, and Adjectives. Cambridge University Press, Cambridge. Belletti, A., 1988. The case of unaccusatives. Linguistic Inquiry 19, 1–34. Borer, H., 2003. Exo-skeletal vs. endo-skeletal explanations: syntactic projections and the lexicon. In: Moore, J., Polinsky, M. (Eds.), The Nature of Explanation in Linguistic Theory. CSLI Publications, Stanford, CA, pp. 31–67. Embick, D., 1997. Voice and the Interfaces of Syntax. Doctoral Dissertation, University of Pennsylvania, Philadelphia. Grimshaw, J., 1990. Argument Structure. MIT Press, Cambridge, MA. Haaften, T., et al., 1985. Nominalisaties in het Nederlands. Glot 8, 67–104. Hale, K., Keyser, S.J., 1993. On argument structure and the lexical expression of syntactic relations. In: Hale, K., Keyser, S.J. (Eds.), The View from Building 20, MIT Press, Cambridge, MA, pp. 53–109. Hale, K., Keyser, S.J., 1998. Bound features, Merge and transitivity alternations. Ms. MIT Press, Cambridge, MA. Hale, K., Keyser, S.J., 2000. Conflation. In: Cuadernos de Linguistica del Instituto Universitaria Ortega y Gasset 7, pp. 39–76. Halle, M., Marantz, A., 1993. Distributed morphology and the pieces of inflection. In: Hale, K., Keyser, S.J. (Eds.), The View from Building 20. MIT Press, Cambridge, MA, pp. 111–176. Harley, H., Noyer, R., 1998. Mixed nominalizations, short verb movement, and object shift in English. In: Tamanji, P.N., Kiyomo, K. (Eds.), Proceedings of NELS 28. GLSA, University of Massachusetts, Amherst, pp. 143–157. Harley, H., Noyer, R., 1999. State-of-the-article: distributed morphology. Glot 4 (4), 3–9. Higginbotham, J., 1983. Logical form, binding and nominals. Linguistic Inquiry 14, 395–420. Hoekstra, J., 1997. The Syntax of Infinitives in Frisian. Doctoral Dissertation, University of Groningen, Fryske Academy, Ljouwert. Hoekstra, T., Wehrmann, P., 1985. De nominale infinitief. Glot 8, 257–275. Marantz, A., 1997. No escape from syntax: don’t try morphological analysis in the privacy of your own Lexicon. In: Dimitriadis, A., et al. (Eds.), Proceedings of the 21st Annual Penn Linguistics Colloquium. Penn Working Papers in Linguistics 4 (2), pp. 201–225. Marantz, A., 2000. Case and licensing. In: Reuland, E. (Ed.), Arguments and Case: Explaining Burzio’s Generalization. John Benjamins, Amsterdam. Marelj, M., 2004. Middles and Argument Structure across Languages. Doctoral Dissertation. LOT series, Utrecht University. Mutaka, N.M., Safir, K., 2008. Kinande anaphora sketch. www.africananaphora.rutgers.edu/downloads/./KinandeAS.pdf. Reinhart, T., 2002. The Theta system – an overview. Theoretical Linguistics 28, 3. Reinhart, T., Siloni, T., 2005. The lexicon-syntax parameter: reflexivization and other arity operations. Linguistic Inquiry 389–436. Reuland, E., 1983. Governing –ing. Linguistic Inquiry 14 (1), 101–136. Reuland, E., 1988. Relating morphological and syntactic structure. In: Everaert, M., et al. (Eds.), Morphology and Modularity. Foris Publications, Dordrecht. Reuland, E., 2011. Anaphora and Language Design. MIT Press, Cambridge, MA. Schoorlemmer, M., 2001. Dutch nominalised infinitives as non-identical twins. Ms. UiL OTS, Utrecht University. Shen, Y., 1985. The Structure of Action in the Short Narrative Text. Unpublished PhD Dissertation, Tel Aviv University. Vinokurova, N., 2005. Lexical Categories and Argument Structure: A Study with Reference to Sakha. Doctoral Dissertation. LOT series, Utrecht University.