Available online at www.sciencedirect.com

ScienceDirect

Adjectival predicators and approaches to complementrealisation

www.elsevier.com/locate/linguaLingua 140 (2014) 83--99

Tor Arne Haugen *

Ivar Aasen Institute for Language and Literature, Volda University College, P.O. Box 500, NO-6101 Volda, Norway

Received 19 February 2013; received in revised form 12 December 2013; accepted 17 December 2013Available online 24 January 2014

Abstract

Most research on complement realisation has been concerned with verbs, and a central assumption, especially in theoreticallyoriented approaches, has been that complement realisation, often referred to as argument realisation, is predictable from the structure ofevents, i.e. (a part of) the semantic structure of the verb. I refer to this assumption as the Predictability Thesis. Other valency carriers thanverbs also need to be accounted for, and the article investigates the neglected area of adjectival valency. Two major versions of thePredictability Thesis are assessed: the aspectual approach and the causal approach. The findings support the view that valency is notpredictable from these facets of event structure. Rather, the view that valency belongs to the idiosyncratic aspects of language issupported; it is necessary to specify both the semantic structure and the realisation of complements in the lexicon.© 2013 Elsevier B.V. All rights reserved.

Keywords: Complement realisation; Adjectives; Predictability; Aspectual approach; Causal approach

1. Introduction

Most research on complement realisation has been based on verbs, and different approaches have been concernedwith different facets of the event structure of verbs, see Levin and Hovav (2005) for a comprehensive survey. A basicpremise has been the hypothesis that the realisation of complements is predictable from the structure of events, i.e. (a partof) the semantic structure of the verb. I refer to this assumption as the Predictability Thesis. Given the central role of theverb in the clause, it is hardly surprising that most models of complement realisation have been developed on the basis ofverbal valency. There are also other valency carriers that need to be considered, however, and the present articleinvestigates the neglected area of adjectival valency.

The investigation is based on a corpus study of 181 polyvalent adjectives in Norwegian, in which 956 valencyconstructions, i.e. a particular adjective in a particular pattern, were recorded in a database. A polyvalent adjective isunderstood as an adjective taking at least two complements denoting event participants in at least one of its valencypatterns. The focus is restricted to adjectives in predicative function.1 The term valency construction has been adoptedfrom Herbst and Schüller (2008), and a construction is to be understood as a conventionalised pairing of form andmeaning, as the term is used in the construction grammars of Goldberg (1995, 2006) and Croft (2001). It should be noted,

* Tel.: +47 97619123.E-mail addresses: tor.arne.haugen@hivolda.no, tahaugen@gmail.com.

1 The different possibilities for valency realisation between attributive and predicative adjectives constitute interesting problems in their ownright, see Haugen (2013) for discussion. Pre-head attributive adjectives do normally not take complements in Norwegian, but this will not befurther discussed here.

0024-3841/$ -- see front matter © 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.lingua.2013.12.007

T.A. Haugen / Lingua 140 (2014) 83--9984

however, that the notion of valency construction refers to a specific predicator in a specific valency pattern. Hence,valency constructions are more specific than Goldberg’s (1995) argument structure constructions.

Like verbs, adjectival predicators enter into constructions expressing different event structures, and when it comes togoverning the basic structure of the clause, adjectives can play much the same role as the class of verbal predicators.Consider the following examples from Norwegian:

(1)

2 Thefrequen

a.

*

*

verb fot than b

Han

rbli ‘remli ‘beco

blir

ain’ is also cme’ and vær

endelig

lassified ase ‘be’. We

kvitt

copular will retu

problemet

He becomes finally rid.of problem.DEF ‘Finally, he got rid of the problem’

in the Norwegirn to this in sec

b.

Han blir endelig kvitt He becomes finally rid.of

The adjective kvitt ‘rid of’ obligatorily takes a second complement, in this case the NP problemet ‘the problem’, in additionto the complement in the subject slot. Complements in this second slot will, for easy reference, be referred to as ‘‘objects’’,but no claim is made that they are equivalent to the objects of verbs. ‘‘Subject’’ and ‘‘object’’ should simply be understoodas labels for the different slots. Norwegian has two basic copulas used in constructions with predicative adjectives: være‘be’, which is mostly used in stative contexts, and the inchoative counterpart bli ‘become’, exemplified in (1), which ispredominantly used in non-stative contexts.2 The choice of copula, however, does not alter the valency features of thepredicator in this case:

(2)

a. Han er endelig kvitt problemet He is finally rid.of problem.DEF ‘He is finally rid of the problem’

b.

Han er endelig kvitt He is finally rid.of

Hence, the adjective is the part of the predicator that seems to decide the basic structure of the clause. This observation isin accordance with Heltoft (1995:220), who analysed predicators in the form of adjectives in combination with copularverbs in Danish. He concludes that, ‘‘[i]t is the combinatorial potential of the adjective that determines the number andsyntactic category of the nuclear participants of the clause’’. For example, the adjective kvitt ‘rid of’ determines thecomplements of constructions as in (1) and (2).

In Norwegian, adjectives can take basically the same complement types as can verbs:

(3)

a. Han er redd hunden He is afraid dog.DEF ‘He is afraid of the dog’

b.

Ho er klar over problemet She is aware over problem.DEF ‘She is aware of the problem’

c.

Dei er glade at sumaren kjem

at

snart

They are glad that summer.DEF comes soon ‘They are glad that summer is here soon’

n Referion 4.1.

d.

Seieren er verdt å kjempe for Victory.DEF is worth to fight for ‘The victory is worth fighting for’

In (3a), the adjective redd ‘afraid’ takes an NP object-complement, in (3b) klar ‘aware’ takes a prepositional complement,glad ‘glad’ takes a that-clause in (3c), whereas verdt ‘worth’ takes an infinitive in the object slot in (3d).

Like verbs, adjectives do seem to determine both the number and the types of complements with which they occurin a clause. This means that models of complement realisation also need to account for adjectival valency, and a firststep towards assessing predictability in the complement realisation of adjectives is to investigate whether facets ofevent structure that have been hypothesised to play a role in the complement realisation of verbs, also play a role for

ence Grammar (Faarlund et al., 1997), but this verb is much less

T.A. Haugen / Lingua 140 (2014) 83--99 85

adjectival predicators. Hence, this investigation is also a contribution to an overall assessment of the PredictabilityThesis as such.

A brief overview of the Predictability Thesis is given in section 3, and two major versions of this thesis, the aspectualapproach and the causal approach, are assessed against data from adjectival valency in sections 4 and 5, respectively. It isargued that complement realisation should be considered item-specific knowledge, and a brief account of this view and itsconsequences for grammatical theory is given in section 6. In the next section we start with an overview of the materialinvestigated.

2. Data

In the research on complement realisation the emphasis has tended to be on theorising rather than on large scaleempirical investigations, cf. the survey in Levin and Hovav (2005). As Faulhaber (2011:296) argues, ‘‘[i]t appears to besymptomatic for research on linking that the claims that are made are often based on a number of handpicked exampleswhich clearly indicate a strong relationship between form and meaning but rarely on large-scale samples which areselected without reference to their syntactic properties’’. There is nothing wrong with theorising per se, but it should, ofcourse, be based on a solid empirical foundation. Faulhaber also argues that research on complement realisation hastended to focus on NP complements only, and that prepositional complements and complements in the form of infinitivesand different kinds of clauses have often been neglected. For this reason, too, more large-scale empirical investigationsare needed in this area, and all kinds of valency carriers and all complement types should be included. Of course, otherlanguages than English also need to be investigated.

As already mentioned, the data analysed here come from a corpus-based investigation of 181 polyvalent adjectivesand their valency patterns in Norwegian. It is well known that Norwegian has two written standards, Bokmål and Nynorsk,which were both established around the turn of the twentieth century. Both written standards are represented in thematerial, which comes from one lexicographical corpus of Bokmål (LBC) and a corresponding corpus of Nynorsk (LNC). Inaddition, a large corpus based on Norwegian Web pages has been applied (NWC).3

There are considerable frequency differences between the various adjectives, but overall, constructions withpolyvalent adjectives are relatively infrequent. For this reason, it was necessary to base the investigation on writtencorpora; sufficiently large speech corpora are not at hand for Norwegian. There are some polyvalent adjectives that areconventionally used in one of the two written standards only, but most of the adjectives investigated are used in bothBokmål and Nynorsk. No attempt is made to treat Bokmål and Nynorsk separately; the two written standards arecodifications of the same language. The abbreviations used to refer to the different corpora above have been used to markauthentic examples, whereas unmarked examples are self-constructed. In the next section we will turn to the thesis ofpredictability in complement realisation.

3. The Predictability Thesis

The Predictability Thesis, the assumption that complement realisation is predictable from rules based on the semanticstructure of the valency carrier, has been advocated in approaches to valency connected to different versions ofgenerative grammar; see Levin and Hovav (2005) for a comprehensive survey. Chomsky (1986:87) suggests that, ‘‘if averb (or other head) s-selects a semantic category C, then it c-selects a syntactic category that is the ‘canonical structuralrealization of C’.’’ Levin (1993:5) claims that, ‘‘[p]resumably, predictions about verb behavior are feasible becauseparticular syntactic properties are associated with verbs of a certain semantic type’’, and Pinker (1989:62, emphasis inoriginal) argues that, ‘‘[s]yntactic argument structures of verbs are predictable from their semantic structures, via theapplication of linking rules’’.

The Predictability Thesis is not necessarily restricted to individual languages; two major proposals have been maderegarding the universality of complement realisation (Rosen, 1999), viz. Perlmutter and Postal’s (1984) UniversalAlignment Hypothesis and Baker’s (1988) Uniformity of Theta Assignment Hypothesis. The former was formulated in theframework of Relational Grammar and states that, ‘‘[t]here exist principles of universal grammar which predict the initialrelation borne by each nominal in a given clause from the meaning of the clause’’ (Perlmutter and Postal, 1984:97). A GBversion of this idea is Baker’s (1988:46) Uniformity of Theta Assignment Hypothesis: ‘‘Identical thematic relationshipsbetween items are represented by identical structural relationships between those items at the level of D-structure’’. Thismeans that there is a one-to-one relationship between semantic complements and syntactic positions, and Rosen(1999:6) interprets this as follows: ‘‘Universal alignment predicts identical mappings of arguments into syntax acrossverbs and languages’’. Hence, this very general idea reflects the Predictability Thesis on a universal level, and the

3 More details about the corpora are found in Haugen (2012), on which this article is partly based.

T.A. Haugen / Lingua 140 (2014) 83--9986

hypotheses of universal linking between the semantics and the realisation of complements reflect the strong tendencytowards building far-ranging theories in this area of research.

Perhaps the most basic approach to complement realisation is the semantic-role approach, where each predicator isassociated with an inventory of roles, one for each of the participants in the event denoted. If predictability is assumed, therealisation of complements is hypothesised to follow from these roles. The semantic-role approach is applicable also toadjectival predicators, although there are few studies that attempt to formulate semantic roles for adjectives. One large-scale empirical investigation of English adjectives is Herbst (1983), who does in fact formulate such roles. He shows thatsemantic roles are useful to compare the semantic structures also of adjectival predicators, and he also considers therelationship between the semantic structure of adjectives and their complement realisation. Although he does identifysome correlations between the semantic structure of adjectives and the valency patterns in which they occur, a 1:1correspondence between complements in specific semantic functions and semantically-defined adjective groups is foundonly in three out of the 35 groups he defines. Haugen (2012) also explicates semantic roles for adjectives (in Norwegian),and finds that most roles have multiple realisations across different predicators. Hence, both these large-scaleinvestigations of adjectival valency point in the same direction: It seems to be very difficult to predict complementrealisation based on the semantic structure of adjectives.

Recently, Boas (2003) and Faulhaber (2011) have challenged predictability in complement realisation also in Englishverbs. Faulhaber compares 87 verbs distributed over 22 semantic groups and concludes that ‘‘[i]f semantic determinismreflected language reality then the syntactic differences between these verbs should be retraceable to their meaning. Fora considerable number of restrictions identified this was simply not possible’’ (p. 299). Hence, also this large-scaleempirical study suggests that complement realisation should be considered to be item-specific knowledge. Faulhaberalso compares verbs on the basis of the semantic roles of their complements. Here, we will focus on two other mainversions of the Predictability Thesis, namely the aspectual approach and the causal approach. These approachesrepresent different conceptualisations of the events that valency carriers and their complements denote.

‘‘Event structure’’ is commonly used to refer to the representation of the part of the semantic structure of a predicatorwhich is assumed to determine the realisation of complements. According to Levin and Hovav (2005:78), this term,‘‘reflects the consensus that such representations encode properties of events’’, but there is no consensus as to howevents should be represented nor as to what parts of an event structure that are relevant to predict the realisation ofcomplements. The term ‘‘event’’ is used to refer to all aspectual types of events, also to stative events; cf. Croft (2012:34).As discussed above, the semantic-role approach is perhaps the most basic approach to event structure. Two more recentapproaches to event structure, which have figured prominently in the research on complement realisation, are theaspectual approach and the causal approach. These approaches represent different conceptualisations of events, andthey attempt to account for complement realisation by linking it to aspectual and to causal facets of events, respectively.Hence, they can be seen as alternative versions of the Predictability Thesis.

In my view, the aspectual- and the causal approaches can be seen as alternative ways of defining semantic roles; theaspectual and causal roles of events are also filled by participants in some way, as starting points and ending points oftemporal and causal chains, respectively. Hence, the participants in the events described are central also in theseapproaches. Clearly, the main facets of events that are highlighted here reflect the fact that the research in this area haspredominantly been carried out on verbs, and this raises an important question: To what extent are these facets of eventstructure relevant in predicting the complement realisation of adjectives? This is the question that will be explored in thesections that follow.

4. The aspectual approach to complement realisation

The aspectual approach to event structure was not originally developed to account for complement realisation (Levinand Hovav, 2005:87). From the outset, aspectual classification was concerned with verbs, and Vendler’s (1957) sub-division of verbs into four aspectual classes is probably the most influential (cf. Rosen, 1999:3; Levin and Hovav, 2005:88;Croft, 2012:33ff). Vendler’s classes can be distinguished through three pairs of semantic features (Croft, 2012:35;following Mourelatos, 1981): STATIVE/DYNAMIC, DURATIVE/PUNCTUAL, and TELIC/ATELIC4:

(4)

4 Upp

States:

er case letters are used

stative

to refer to

durative

components

atelic

Activities: dynamic durative atelic Achievements: dynamic punctual telic Accomplishments dynamic durative telic

of semantic structure.

T.A. Haugen / Lingua 140 (2014) 83--99 87

Rosen (1999:3) gives the following examples of these classes:

(5)

STATES

a.

Terry knows the answer b. Terry resembles his brother

(6)

ACTIVITIES

a.

Terry walked for an hour b. Terry is driving the car

(7)

ACHIEVEMENTS

a.

Terry reached the summit in 15 minutes b. The vase broke

(8)

ACCOMPLISHMENTS

a.

Terry built five houses in two months b. The child is drawing a circle

Frawley (1992:294, emphasis in original) defines aspect as, ‘‘the way that an event is distributed through the time frame inwhich the event occurs’’, and it is an important point that this definition does not specifically refer to verbs. It has beenfrequently demonstrated in the literature that aspect cannot be ascribed to the verb only; also ‘‘characteristics of theobject, adjuncts, and other materials in the clause contribute to the event type of the entire clause’’ (Rosen, 1999:4). Acommon test to discern between atelic statives and activities on the one hand and telic achievements and accomplish-ments on the other is to check for compatibility with DURATIVE adverbials with for, indicating atelicity, and CONTAINERadverbials with in, indicating telicity (cf. Croft, 2012:35f; Arche, 2006:63f). Based on these tests, the following examples,taken from Rosen (1999:4), illustrate that various clause components other than the verb can change the aspect of theclause:

(9)

ADDITION OF DIRECT OBJECT

a.

Bill ran for 5 minutes/*in 5 minutes (activity) b. Bill ran the mile *for 5 minutes/in 5 minutes (accomplishment)

(10)

SPECIFICITY OF OBJECT

a.

Bill wrote letters for an hour/*in an hour (activity) b. Bill wrote the letter *for an hour/in an hour (accomplishment)

(11)

COUNT/MASS OBJECT

a.

Bill drank coffee for an hour/*in an hour (activity) b. Bill drank a cup of coffee *for an hour/in an hour (accomplishment)

(12)

VERB PARTICLE

a.

Terry thought for an hour/*in an hour (activity) b. Terry thought up an answer in an hour/*for an hour (accomplishment)

(13)

RESULTATIVE

a.

Terry ran for an hour/*in an hour (activity) b. Terry ran us ragged in an hour/*for an hour (accomplishment)

These examples indicate that aspect is compositional; it cannot be due to the aspectual class of the verb only.After this brief characterisation of aspect, we will take a look at how aspect has been claimed to play a role in

complement realisation. According to Levin and Hovav (2005:96), the idea that there is a link between aspect and therealisation of complements can be traced back to Hopper and Thompson (1980) and their seminal paper ontransitivity. According to Hopper and Thompson, telicity and punctuality are among the factors that contributeto transitivity. They see transitivity as a gradual phenomenon connected to clauses; clauses have different degreesof transitivity, and telicity and punctuality are factors that increase it. It is an important point that transitivity isconnected to clauses and not only to verbs per se, because this is a feature that transitivity shares with aspect, as wesaw in (9)--(13).

Aspectual approaches to complement realisation are rather limited in scope. The main proposal is that certain kinds ofdirect objects are associated with telicity, or with what Dowty (1991) calls incremental themes; cf. Levin and Hovav(2005:109). Borer (2005:50) argues that, ‘‘only direct arguments interact with event structure’’, and Tenny (1994:10f)argues that only direct internal complements, by which she means NP objects, can measure out events, and, hence,realise the aspectual feature in question. Verbs taking other kinds of complements do not fall under the Measuring-OutConstraint. Tenny’s notion of ‘‘measuring out’’ is similar to Dowty’s ‘‘incremental theme’’, and we will now take a closer

T.A. Haugen / Lingua 140 (2014) 83--9988

look at these notions. For Dowty, an incremental theme is one kind of participant role.5 An incremental theme is ‘‘an NPthat can determine the aspect of the sentence, since the parts of the event correspond to parts of the NP referent that areaffected by the action; the event is ‘complete’ only if all parts of the NP referent are affected (or effected)’’ (Dowty,1991:588). Consider the following examples:

(14)

5 An imseveral fbetween

6 The t

a.

portantactors

the noterm ‘‘pr

Han

differethat chaion of ‘edicate

åt

nce beracte‘meas’’ here

pølsa

He ate sausage.DEF ‘He ate the sausage’

tween the approrise prototypicauring out’’ and d

refers to verb

b.

Han åt He ate ‘He ate’

In (14a), the direct object pølsa ‘the sausage’ is the realisation of an incremental theme complement of the verb ete ‘eat’.The theme complement is incremental in that pølsa is gradually consumed as the eating event progresses. The parts ofthe event of eating it correspond to the parts of pølsa ‘the sausage’, and when it is fully consumed, the event reaches itsendpoint. Hence, the direct object makes the construction telic, which is clearly demonstrated by contrasting it with theexample in (14b), which is atelic. In Tenny’s (1994) terms, the direct object measures out the event in that it brings theevent to an endpoint when it is fully consumed. This kind of measuring out is then extended to other kinds of predicates6:

(15)

a. Dei gjekk Besseggen They walked Besseggen ‘They walked Besseggen’

achl instirec+ no

b.

Vinden tørka kleda Wind.DEF dried clothes.DEF ‘The wind dried the clothes’

In (15a), we have what Tenny (1994:17) calls a ‘‘path object’’; in our case the path is the mountain ridge Besseggen, andthis path measures out the event in that the progress along the path is the measure for the progress of the event, and theendpoint of the event is the endpoint of the path. In (15b), we have an inchoative verb, whose patient complement Tennycharacterises as the measure of the event. In this case, the measure is not connected to the extension of the object, but tothe progress of the object on the scale of DRYNESS denoted by the verb. As Tenny (1994:18) and Levin and Hovav(2005:95) argue, the common feature of these verbs is that they involve a scale, and that progress through the eventcorresponds to progress along this scale. In (14a), the scale is found in the physical extension of the direct object, whereasin (15a), the scale is found as a path with an endpoint within the direct object. It is the endpoint of the scale that gives telicityto the construction (Tenny, 1994; Levin and Hovav, 2005). This is why it has been suggested that the notion of incrementaltheme should be reserved for the scale which functions as a measure of the event, and that the complement that isrealised as a direct object is the complement about which a progress on the scale is predicated (Levin and Hovav,2005:99; citing Kennedy and Levin, 2001).

Tenny (1994:11) argues that, ‘‘[d]irect internal arguments are the only overt arguments which can ‘measure out theevent’’’, but this claim is challenged by evidence presented by Dowty (1991:570f), who argues that examples like in(16a)--(16b) have incremental theme subjects, whereas the one in (16c) has an incremental theme PP:

(16)

a. John entered the icy water (very slowly) b. The crowd exited the auditorium (in 21 minutes) c. She walked across the desert in a week

Hence, the notion of incremental theme cannot be used to predict that a certain complement will be realised as a directobject. Jackendoff (1996) argues that it is AFFECTEDNESS and not ‘‘measuring out’’ that is most central to its realisationas a direct object, and he demonstrates that not all affected objects are incremental themes (p. 312):

(17)

John chewed/kneaded/jiggled/spun the loaf of bread for/*in an hour

es of Dowty (1991) and Tenny (1994) is that for Dowty, the notion incremental theme is one ofances of the role patient, whereas Tenny makes more direct claims about the correlationt objects.n-subject complements.

T.A. Haugen / Lingua 140 (2014) 83--99 89

With these verbs, the direct object the loaf of bread is affected, but it does not measure out the event: the event is atelic. Itis clear that not all direct objects are incremental themes; rather, this is verb-specific, but in some cases there is acorrelation between the presence of a direct object and telicity. Hence, the relationship between aspect and complementrealisation is correlational rather than implicational.

4.1. The aspectual approach and adjectival predicators

After this brief outline of the aspectual approach to complement realisation, its main claims, and some of the problemswith which it is concerned, we will now turn our attention to adjectival predicators. To what extent is the aspectualapproach to event structure and complement realisation applicable also to adjectives? To answer this, we have to take acloser look at the composition of complex adjectival predicators. At clause level, adjectives combine with various verbs toform complex predicators, but in the present investigation, the focus will be restricted to combinations with full-fledgedcopular verbs. The goal of this investigation is to reveal the valency properties of the adjective category, and predicatorscomposed of a copular verb and an adjective show more valency variation than combinations with other verbs; cf.Daugaard (2002:107f) and Faarlund et al. (1997:734).

The full-fledged copulas have very schematic meanings. The three Norwegian verbs that are regarded as copulasby Faarlund et al. are være (Nynorsk vere) ‘be’, bli (Nynorsk also verte) ‘become’, and forbli ‘remain’; see alsoDaugaard (2002:115), who concludes that these three are also the full-fledged copulas in Danish. The semanticcontributions of these verbs in combination with adjectives seem to be restricted to expressing tense and aspect. AsDaugaard points out, the full-fledged copulas complement each other aspectually (Daugaard, 2002). Være ‘be’ andforbli ‘remain’ are both stative. Bli ‘become’, on the other hand, is the inchoative counterpart of være and expressesthat the state denoted by the predicator comes into being; see also Faarlund et al. (1997:734). This is the aspectualdifference we will focus on here:

(18)

7 We w

a.

ill retur

Huset

n to some exc

er

eptio

blått

House.DEF is blue ‘The house is blue’

ns fro

b.

Huset blir

m

blått

House.DEF becomes blue ‘The house becomes blue’

The basic aspect of constructions with adjectival predicators in combination with være, as in (18a), is stative, whereasconstructions with bli ‘become’ are inchoative, as in (18b).7

As pointed out in our discussion above, telicity and the notion of incremental theme have figured prominently inaspectual approaches to complement realisation. Therefore, in an assessment of the extent to which the aspectualapproach is applicable to adjectives, the following questions are a natural starting point: Are constructions with adjectivalpredicators telic, and do they include incremental themes? We will first consider telicity.

The difference between the constructions in (18) clearly involves a difference in telicity. (18a) is atelic, whereas (18b) istelic. The difference between være ‘be’ and bli ‘become’ can also be demonstrated with the distinction betweenDURATIVE and CONTAINER adverbials; see (9)--(13). DURATIVE for adverbials are expressed in Norwegian with thepreposition i, whereas CONTAINER in adverbials, are expressed with på:

(19)

a. Jeg var sur i en

thes

uke (LBC)

I was sour for a week ‘I was grumpy for a week’

e basic case

b. ?

Jeg var sur på en uke I was sour in a week

c. ?

Jeg ble sur i en uke I became sour for a week

d.

Jeg ble sur på en uke I became sour in a week ‘I turned grumpy in a week’

s in (27).

T.A. Haugen / Lingua 140 (2014) 83--9990

I ‘for’ adverbials are much more conventional with være, whereas på ‘in’ adverbials are combined with bli. Hence, theconstructions with være are most easily interpreted as atelic, whereas the constructions with bli are interpreted as telic.8

We can conclude from this that in Norwegian, the choice between the copulas være ‘be’ and bli ‘become’ is normallyassociated with a change in aspect, more specifically with a change in telicity. A further aspectual difference found with bliconstructions is between durative events (20a) and punctual events (20b):

(20)

8 An al

a.

ternativ

jeg

e read

ble

ing of (19c

god

) is an it

i

erat

matte (NWC)

I became good in maths ‘I became good at maths’

ive reading where

b.

Skjermen ble svart (LBC) Screen.DEF became black ‘The screen turned black’

Hence, according to Vendler’s classification in (4), (20a) is an accomplishment, whereas (20b) is an achievement.We will now turn to our second question: Do constructions with adjectival predicators express incremental themes? We

recall from our discussion above that incremental themes are closely associated with telicity, and, hence, we would expectto find them in constructions with the telic bli ‘become’, as in (20). We also recall from our discussion of the examples in(15) that the notion of measuring out is associated with a scale; e.g. a scale of DRYNESS in (15b). Adjectives are alsofrequently associated with a SCALE, and it is the conceptualisation of a change on a scale that makes constructions likethose in (20) telic. The endpoint of the event is reached when the subject of the construction reaches the point on the scalewhere the property denoted by the adjective is reached. For example, the endpoint of the change in (20a) is reached whenthe subject hun ‘she’ reaches a certain point on the scale associated with the adjective god ‘good’, specified as pertainingto MATHEMATICAL SKILLS by the complementation i matte ‘in maths’; the construction expresses that her mathematicalskills have changed and that they have improved. Hence, this is parallel to the construction in (15b), where the directobject is the complement which undergoes a change on a scale of DRYNESS implied by the verb. In that case, the directobject is the complement of which the change on the scale is predicated. With the adjectival predicator, however, theincremental theme is associated with the subject, not with a direct object.

Now, an interesting question is whether this changes if the adjective takes an NP object. Consider the following examples:

(21)

a. Jeg ble lei håret (NWC) I became tired hair.DEF ‘I became tired of my hair’

b.

jeg ble redd denne læreren (NWC)

the sub

I

became afraid this teacher.DEF ‘I became afraid of this teacher’

je

c.

Tittelen ble lik tittelen på plakaten (LBC) Title.DEF became like title.DEF on poster.DEF ‘The title became like the title on the poster’

ct repe

d.

Kroppen hans blir verdt millioner (NWC) Body.DEF his becomes worth millions ‘His body will be worth millions’

In these cases as well, the changes taking place are predicated about the subjects, not about the objects. I conclude fromthis that in adjectives, a change on the scale associated with the adjective is always predicated about the subject, and thismakes incremental theme a poor candidate for predicting complement realisation. If we change the copula into være ‘be’,we get stative constructions, but this has no consequences for complement realisation:

(22)

a. Jeg var lei håret I was tired hair.DEF ‘I was tired of my hair’

b.

Jeg var redd denne læreren I was afraid this teacher.DEF ‘I was afraid of this teacher’

atedly turns grumpy for a week.

T.A. Haugen / Lingua 140 (2014) 83--99 91

c.

Tittelen was lik tittelen på plakaten Title.DEF was like title.DEF on poster.DEF ‘The title was like the title on the poster’

d.

Kroppen hans er verdt millioner Body.DEF his is worth millions ‘His body is worth millions’

The minimal pairs in (21)--(22), differing only in the choice of copula, are also a challenge for a claim made by Arche(2006:107), that aspect is purely syntactic: ‘‘I will defend that there are no inner aspect properties decided from the lexiconbut it is the elements present in the syntactic structure that gives [sic] the aspectual nature of the construction.’’ It seemsclear, however, that there is no syntactic difference between the constructions in (21)--(22). Rather, the two copular verbsexpress different aspects. Nevertheless, let us have a look at the background for Arche’s claim. A central factor in thediscussion is the distinction between individual-level and stage-level predicates, which, according to Arche, is due toCarlson (1977). The former apply to individuals (23a), whereas the latter apply to stages or happenings in whichindividuals participate (23b) (Arche, 2006:6):

(23)

a. John is a mammal b. John is in Los Angeles

In Spanish, adjectival predicators with the copula ser ‘be’ yield individual-level predicates that are normally taken to bestative. Arche observes, however, that there are certain adjectives in Spanish, like cruel ‘cruel’, amable ‘kind’ andmezquino ‘mean’ that can be used with the copula ser (indicating that they are individual-level), but that can still bedynamic. The following Norwegian examples can also illustrate this:

(24)

a. Han er snill (LBC) He is kind ‘He is kind’

b.

Han er snill mot deg (LBC) He is kind against you ‘He is kind to you’

In (24a) the combination of være ‘be’ and snill ‘kind’ yields a stative predicate, whereas the construction in (24b), on theother hand, can be interpreted as a Vendlerian activity, i.e. that he is acting in a kind-hearted way. Arche argues thatconstructions like those in (24b) imply that the subject performs some action directed against the participant denoted bythe PP, although this action is not specified. Arche’s conclusion is that not all individual-level predicates are statives, andthat the non-stativity of such predicates correlates with the presence of what she calls relational PPs, like mot deg ‘to you’in (24b); she argues that the difference between pairs like those in (24) suggests that the aspectual difference is to beascribed to the PP and the construction as a whole and not to the copula. This seems like a very reasonable analysis. Shealso suggests, however, that the difference in (24) is evidence that aspect is purely syntactic.

It is clear that the aspectual difference between (24a) and (24b) is not due to the copular verb, and hence that aspect ispartly determined by the composition of the construction as a whole. This does not mean, however, that there are noaspectual differences between copulas per se. For example, it seems clear to me that the basic aspectual meaning of thecopula være ‘be’ (25a) is different from the basic aspectual meaning of bli ‘become’ (25b), and hence that aspect is bothconnected to the copula per se and constructional:

(25)

a. Jeg er redd hunder (NWC) I am afraid dogs ‘I am afraid of dogs’

b.

Jeg blir redd hunder I become afraid dogs ‘I become afraid of dogs’

In (24b), for example, it seems that the basic aspect of være ‘be’ is overridden when it combines with mot deg ‘to you’ inthat construction. As Croft (2012:31) argues, ‘‘[a]spect is manifested both grammatically and lexically.’’ What theaspectual difference between the examples in (24) demonstrates, however, is that it is too simplistic to regard allconstructions with the copula være + Adj as stative in Norwegian as well.

T.A. Haugen / Lingua 140 (2014) 83--9992

An interesting point that has emerged is that aspect is not only dependent upon the choice of copula. Some adjectivesarguably take AGENT-like subjects, and although these adjectives, per se, do not specify any action, these constructionsas wholes are clearly dynamic and are to be classified as Vendlerian activities:

(26)

a. Jeg er veldig forsiktig med å gå ut (NWC) I am very careful with to go out ‘I am very careful about going out’

b.

han var vel opptatt med å betale ut penger (NWC) he was probably busy with to pay out money ‘He was probably busy paying out money’

It is also possible for constructions with bli ‘become’ to be given a stative reading; Faarlund et al. (1997:735) give theexample in (27a), where the construction contains a continuative adverbial of time. In this example, the predicative is anNP, but a stative reading is also possible with an adjectival predicative as in (27b). Furthermore, constructions with bli canalso express a stative aspect in combination with present participles as in (27c):

(27)

a. Han ble en stakkar hele sitt liv He remained a miserable.wretch all his life ‘He remained a miserable wretch all his life’

b.

Hele vinteren ble kald Whole winter.DEF remained cold ‘It remained cold for the whole winter’

c.

Vi blir sittende We remain sitting ‘We remain seated’

Even though it is not the case that constructions with være are always stative and that constructions with bli are alwaysinchoative and telic, this aspectual distinction is very often found between minimal pairs, as in (25), where only the copulais changed. Hence, if aspect, and telicity in particular, is important for complement realisation, the expectancy would bethat the choice of copula should influence the range of valency patterns that adjectival predicators occur in. However, thisseems to be the case only to a very limited extent.

Nevertheless, an investigation of the valency patterns of 181 polyvalent adjectives in Norwegian shows that there aresome restrictions on the use of the copulas være ‘be’ and bli ‘become’ with some adjectives. Interestingly, however, in therestrictions that are found, it is, in fact, the use of the inchoative and telic bli that is limited, whereas være can almostalways be used. Thus, to the extent that it makes a difference, telicity would appear to restrict the valency patterns ofadjectival predicators.

There are two main types of restrictions on the choice of copula. First, in the case of some predicators, there is a strongpreference for være rather than bli in all their valency patterns. This applies to a few adjectives with the prefix u-- ‘in--, un--’,like ukjent ‘unknown, unfamiliar’:

(28)

a. han var ukjent med denne saka (LNC) he was unfamiliar with this matter.DEF ‘He was unfamiliar with this matter’

b. ?

han blei ukjent med denne saka he became unfamiliar with this matter.DEF

As Daugaard (2002:253) points out, predicators confined to the stative aspect are non-scalar; they denote either-or states.In addition, the most common change when it comes to knowing something is first to not know something, and then to getto know it. This is the change expressed by (29b), which is fully conventional:

(29)

a. han var kjent med denne saka he was familiar with this matter.DEF ‘He was unfamiliar with this matter’

b.

han blei kjent med denne saka he became familiar with this matter.DEF ‘He became familiar with this matter’

T.A. Haugen / Lingua 140 (2014) 83--99 93

In this case, the change is from the negative to the positive pole. When bli is combined with ukjent, on the other hand, thechange is from the positive to the negative pole, which is less common. It is possible, however, in an example like thefollowing:

(30)

Han blei plutselig ukjent med denne saka då politiet ringde He became suddenly unfamiliar with this matter.DEF when police.DEF called ‘He suddenly became unfamiliar with this matter when the police called’

Hence, this restriction is context-dependent and not absolute, and it is not the case that the choice of copula influences therange of possible valency patterns in any absolute way.

The second main kind of restriction on the use of the copula bli ‘become’ is found with some predicators when they takea subject in the form of an assertive finite clause (cf. Daugaard, 2002:245). Among the 181 adjectives in my database, thisis the case for ansvarlig ‘responsible’; bevisst ‘deliberate’; medviten ‘conscious’; moden ‘mature’; sikker and viss ‘certain’;and takknemlig ‘thankful’:

(31)

a. Det var bevisst at enkelte kostnader ikke var tatt med i søknaden (NWC) It was deliberate that certain costs not were taken with in application.DEF ‘It was deliberate that certain costs were not included in the application’

b. ?

Det ble bevisst at enkelte kostnader ikke var tatt med i søknaden It became deliberate that certain costs not were taken with in application.DEF

c.

at det har vært brukt i tusener av år, er helt sikkert (LBC) that it has been used in thousands of years, is totally certain ‘That it has been used for thousands of years, is certain’

d. ?

at det har vært brukt i tusener av år, ble helt sikkert that it has been used in thousands of years, became totally certain

When they combine with assertive finite-clause subjects, these predicators are used to evaluate processes whose truththe conceptualiser is committed to, and the moment this commitment is made, the processes are also evaluated asconscious (31a) or unconscious, for example. At the moment of utterance, this evaluation cannot change, as the use of bli‘become’ would express (31b). These restrictions on the use of bli, are, however, also dependent on the context, and if thecontext is changed, bli is also possible:

(32)

a. at de t har vært b rukt i tusener av år ble helt s ikkert etter de nye funnene that it has been sed in thousands of years became totally ertain after the new findings.DEF u c‘That it has been used for thousands of years is certain after the new findings’

In other valency patterns, these predicators occur freely with bli:

(33)

a. De har blitt mer bevisst på egen kunnskap (NWC) They have become more conscious on own knowledge ‘They have become more conscious of their own knowledge’

b.

Jeg er nok blitt sikrere på meg selv (NWC) I have probably become more.secure on me self ‘I have probably become more secure in myself ’

The clearest cases of correlations between the choice of copula and possible valency patterns are found in the predicatorpairs enig/uenig and samd/usamd, both meaning ‘agreed/disagreed’; cf. Daugaard’s (2002:250) analysis of the Danishequivalents enig/uenig. These predicators allow bli when they occur with complements with the preposition om ‘about’(34a), but not when they occur with the preposition i ‘in’ (34b). With være, on the other hand, both complement types arepossible, as demonstrated by (34c) and (34d):

(34)

a. Vi ble enige om at de ikke skulle vente (LBC) We became agreed about that they not should wait ‘We agreed that they should not wait’

b. ?

Vi ble enige i at de ikke skulle vente We became agreed in that they not should wait

T.A. Haugen / Lingua 140 (2014) 83--9994

9 The

c.

adjectiv

Vi

e var

er

can a

enige

lso mean

om

‘sensitiv

at

e’, in

den

which

tredje

case it a

verden

lso combine

skal

s with v

få

ære

oppfylt

‘be’.

sine

grunnleggende behov (LBC) We are agreed about that the third world.DEF shall get fulfilled their basic needs ‘We agree that those in the Third World should get their basic needs fulfilled’

d.

Vi er enige i at den tredje verden skal få oppfylt sine grunnleggende behov We are agreed in that the third world.DEF shall get fulfilled their basic needs ‘We agree that those in the Third World should get their basic needs fulfilled’

In this case, the different possible valency patterns correlate with a change in aspect: the combination enig i is used instative contexts only, whereas enig om is used in both stative and inchoative, telic contexts. However, such cases seem tobe very rare, and they do not mean that aspectual properties can be taken to predict complement realisation.

Finally in this section, we will take a look at the only case in my material where an adjective is restricted to a combinationwith the copula bli ‘become’. The case in point is var, which has only been found with bli (Nynorsk also verte) when itmeans ‘aware’9:

(35)

Det var han som vart var henne (LNC) It was he who became aware her ‘He was the one who became aware of her’

As previously mentioned, telicity and incremental themes are the aspectual notions that have figured most prominently inaspectual approaches to complement realisation. The data presented here, however, suggest that these notions do notplay any major role for complement realisation in adjectival predicators. With some exceptions, as in (27), constructionswith the inchoative copula bli ‘become’ are telic, but in most cases adjectival predicators occur in the same valencypatterns when they combine with bli as when they combine with the stative være ‘be’. This finding is in accordance withDaugaard (2002:238), studying Danish, who also concludes that være and bli can be used interchangeably with mostadjectival predicators in all their valency patterns. There are a few cases where the choice of copula restricts valencypatterns, but we have seen that the use of the telic bli is, in fact, more restricted than the use of være; the adjective var‘aware’ represents a rare exception to this generalisation. Furthermore, the predictability of complement realisation fromaspect cannot be based on the evidence of a handful of predicators. There are too many cases that cannot be accountedfor with this approach, and the correlations we have seen between choice of copula and valency patterns do not alter ourmain conclusion, viz. that both copulas can be used interchangeably with most adjectival predicators in all their valencypatterns. Hence, my conclusion is that an aspectual approach to complement realisation faces serious challenges whenencountering data from adjectival valency.

Anticipating the account in the next section of the causal approach to complement realisation, it should be pointed outthat proposals have been put forward to also connect the realisation of subjects to an aspectual notion. Voorst (1988)proposed that both subject- and object realisation are linked to the temporal structure of events. Telic events, i.e.achievements and accomplishments, are limited by a point in time that initiates the event, and a point that terminates it.The initiation point is identified through the ‘‘object of origin or actualization’’ (Voorst, 1988:10), whereas the endpoint isidentified through the ‘‘object of termination’’ (Voorst, 1988). In complement realisation, the object of origin is realised assubject, whereas the object of termination is realised as object, albeit in ‘‘deep structure’’. Except for the notion of ‘‘deepstructure’’, this approach to complement realisation is in fact not that dissimilar from the causal approach, where theinitiators and endpoints of causal chains play a central role in complement realisation; the difference being that the chainbetween initiator (or originator) and endpoint is temporal in the aspectual approach and causal in the causal approach.Levin and Hovav (2005:125ff) also discuss these similarities and point out that ‘‘[t]emporal precedence often correspondsto precedence in the causal chain’’ (Levin and Hovav, 2005). The causal approach to complement realisation is ourconcern in the following section.

5. The causal approach to complement realisation

The causal approach to event structure is an approach that is most closely linked to cognitive linguistics and hasTalmy’s (1976, 1988) work on causation and force dynamics as an important basis: ‘‘Force dynamics is a generalizationover the traditional linguistic notion of ‘causative’: it analyzes ‘causing’ into finer primitives’’ (Talmy, 1988:49). Forcedynamics involves such concepts as, ‘‘the exertion of force, resistance to such exertion and the overcoming of suchresistance, blockage of a force and the removal of such blockage, and so forth’’ (Talmy, 1988). Talmy also extends theforce dynamics involved in physical interaction between entities to psychosocial interactions. In the causal approach to

T.A. Haugen / Lingua 140 (2014) 83--99 95

Fig. 1. Causal chain.

event structure, events are described as causal chains (Croft, 1991, 1998, 2012) or action chains (Langacker, 1991), andcommon to both these approaches is the fact that such chains are divided into segments that relate participants in theevent, and that participants are linked by relationships of causal directionality. The participant starting a causal chain iscalled an initiator, whereas the participant ending it is the endpoint. Based on Talmy’s (1988) causation types, Croft(1991:166f) distinguishes between four kinds of causation, where initiatiors and endpoints are either mental or physical;see also Croft (2012:200):

� P

hysical causation: physical initiator--physical endpoint � V olitional causation: mental initiatior--physical endpoint � A ffective causation: physical initiator--mental endpoint � In ducive causation: mental initiatior--mental endpoint

For example, the clause in (36) can be given the causal chain representation in Fig. 1, adopted from Croft (2012:222).

(36)

Sue broke the coconut for Greg with a hammer

In the causal chain in Fig. 1, we have three different kinds of causations represented: a volitional causation when Sue acts onthe hammer, a physical causation when the hammer acts on the coconut, and an affective causation when the coconut actson Greg (in that its breaking benefits him). Hence, causation is interpreted quite broadly, and what happens between aninitiator and an endpoint can be referred to as a transmission either of force or some kind of concrete or abstract content.

Croft (2012:198) makes an attempt to reconcile the causal approach with the aspectual approach to complementrealisation, but he maintains that, ‘‘[t]he fundamental semantic property that determines participant role ranking forargument realization is the causal structure of events’’. In the causal approach, complement realisation is based on the roleof event participants in the causal chain and on their relations to other participants in the chain: participant roles are rankedaccording to their position in the causal chain profiled by individual predicators. This represents a difference from semantic-role hierarchies, which include semantic roles that never occur with the same predicator; in the causal approach tocomplement realisation, ranking between roles is determined by event structure, ‘‘in the sense of participants in the sameevent’’ (Croft, 2012:177), i.e. in the sense of participants denoted by individual predicators. The causal approach, however,also posits realisation rules that are based on generalisations over different kinds of events (see Croft, 2012:207), i.e. overthe event structures of more than one predicator, so it is not clear to me that the difference between this approach and otherrole hierarchies is necessarily that great. Initiators and endpoints of causal chains are also one kind of participant roles, andthe ranking between these roles is assumed to hold across predicators and to account for complement realisation. Hence,the causal approach is also an attempt to predict complement realisation based on generalisations over event types andacross different predicators. The initiator and the endpoint of the event profiled by the predicator are linked to thegrammatical roles subject and direct object, respectively, whereas the realisation of oblique grammatical roles follows fromtheir position relative to the direct object.

In our example in Fig. 1, the predicator break profiles the causal chain from Sue, which is realised as subject, to thecoconut, which is realised as direct object. The profiled part of the causal chain is indicated by the solid arrows. Theoblique with a hammer is a participant bearing an INSTRUMENTAL role; the preposition with is what Croft (1991:Ch. 5)calls an antecedent oblique marker; antecedent obliques precede the direct object in the causal chain. For this reason,antecedent obliques are also part of the event profiled by the predicator. The oblique for Greg, on the other hand, is asubsequent oblique, which follows the direct object in the causal chain. Subsequent obliques are part of the semanticframe of the event, but are not profiled, as indicated by the stippled arrow. The central prediction made by the causalapproach to complement realisation is that the initiator of the causal chain should be realised as subject, whereas theendpoint of the chain should be realised as direct object.

5.1. The causal approach and adjectival predicators

Turning now to adjectival predicators, there are two basic problems that make it difficult to apply a causal chain analysisto this class. First, as discussed in the previous section, adjectival predicators combined with the copula være ‘be’ arestative in most cases, and causal chains are only applicable to dynamic events. Furthermore, constructions with thedynamic copula bli ‘become’ do not, with most predicators, describe causal relations either:

(37) a. Jeg ble kvitt problemet allerede etter 17 timer (NWC)

T.A. Haugen / Lingua 140 (2014) 83--9996

I

got rid.of problem.DEF already after 17 hours ‘I got rid of the problem already after 17 hours’

b.

Midtperioden ble lik flere av kampene lagene mellom forrige Middle.period.DEF became like several of matches.DEF teams.DEF between last sesong (NWC) season ‘The middle period became similar to several of the matches between the teams last season’

c.

Salgsmessig har hiphopkulturen de siste årene blitt overlegen rocken (NWC) Sales.wise has hiphop-culture.DEF the last years.DEF become superior rock.DEF ‘In terms of sales, the hiphop-culture has in recent years surpassed the rock culture’

d.

Kroppen hans blir verdt millioner (NWC) Body.DEF his becomes worth millions ‘His body will be worth millions’

In (37a), the NP complement problemet ‘the problem’ expresses what the subject got rid of without expressing any causeof this event, and there is no causal relation between either the subject or the object and the properties described by theadjectives lik ‘like’ in (37b), overlegen ‘superior’ in (37c), or by verdt ‘worth’ in (37d) either. Hence, causal relations cannotpredict why these adjectives should be polyvalent. There are, however, some predicators with which constructions with blido express causal relations. Consider the following examples:

(38)

a. jeg ble redd hunden (NWC) I became afraid dog.DEF ‘I became afraid of the dog’

b.

han ble stolt av sin kone (LBC) He became proud of his wife ‘He became proud of his wife’

With these predicators, however, it is often not evident which participant is to be considered the initiator: the subjects or theobject complements of these constructions. To claim a correlation between the causal chain and realisation, we wouldhave to say that the subject is the initiator in these cases, and this might be argued for, based on the necessity of thesubject directing its attention to the participant denoted by the other complement. However, in (38a) it is also the case thatthe object denotes a CAUSE of the feeling of being redd ‘afraid’ and in (38b) that the prepositional complement denotesthe CAUSE of the subject being stolt ‘proud’. Hence, these complements precede the subjects in the causal chain, and thesubjects are mental endpoints. These adjectival predicators impose the same problems on the causal approach tocomplement realisation as do verbs of the kind Croft (1991:251) calls reverse verbs, like hear, receive, suffer and undergo,whose subjects are also endpoints and not initiators in the causal chain. Note that this problem would also apply toaspectual, temporal chains discussed at the end of the previous section.

An additional difficulty with the causal approach, which is especially prominent when dealing with adjectivalpredicators, is that a majority of adjectival predicators do not take NP objects; of the 181 adjectives in my database,only 34 (19 per cent) actually occur with a second NP complement. A large majority of adjectival predicators takeprepositional complements. If prepositional complements are classified as obliques, the causal approach cannotaccount for them, since it only accounts for obliques relative to direct objects. Croft himself (1991:176) also points outthat, ‘‘[o]bligatory obliques, as in look for a book, are a more direct problem for the causal structure analysis. If theverb meaning is just the causal chain between the subject (initiator) and object (endpoint), then an obligatory obliqueis an anomaly’’. However, prepositional complements of the kind we have in (38b) are clearly valency-bound, andneed to be accounted for in a model of complement realisation. In conclusion, the causal approach to complementrealisation does not seem very promising when it comes to predicting the complement realisation of adjectivalpredicators.

If the realisation of complements is not predictable, the following questions arise: How do we account for valency? Andwhat consequences does this have for our model of grammar? These questions will be discussed in the section below.

6. Valency as item-specific knowledge

The evidence presented in the previous sections suggests that the realisation of complements is predictable neitherfrom aspectual nor from causal features. The present investigation only includes Norwegian adjectives, but thePredictability Thesis is also challenged by other large-scale investigations of valency. Herbst (1983), studying English

T.A. Haugen / Lingua 140 (2014) 83--99 97

adjectives, and Faulhaber (2011), studying English verbs, also conclude that it seems very difficult to predict complementrealisations on the basis of the semantic structures of the valency carriers (and their complements).

The question of predictability in complement realisation is fundamental to grammatical theory. Two differentconceptions of grammar can be juxtaposed: grammar as a rule-based, generative system, and grammar as a repository ofmore or less schematic symbolic units (or constructions). If complement realisations are predictable, this speaks in favourof the former view, but if not, grammar looks more like a repository of symbolic units.

The Predictability Thesis can be seen as an attempt to minimise the lexicon: lexical entries should contain as littleinformation as possible, and if the realisation of complements is predictable, this information can be left out. On the otherhand, however, as many of a predicator’s complementation patterns as possible should be predictable from the lexical entry.Boas (2006:134) refers to this as the lexicalist paradox: ‘‘The lexicalist paradox is the situation in which we would like toexplain all of a verb’s argument realization options based on a very minimal lexical entry and a number of generativemechanisms or constructions’’. If complement realisation is fully predictable, only semantic representations of complementsare needed in the lexicon, since realisations follow from general rules. If realisations are item-specific, on the other hand, theyneed to be linked to the semantic representations of the complements in the lexicon. We thus end up with symbolicrepresentations not only of the valency carriers, but also of the complements, and the representations of valency carriers(with complements) end up very much like the constructions of construction grammars, i.e. as (complex) pairings of form andmeaning (cf. Goldberg, 2006; Croft, 2001). In such a model, the basic structure of clauses relies heavily on a vast repository ofstored symbolic units (constructions), since valency carriers need to be stored together with all the patterns in which theyconventionally occur. Hence, valency is based on rich memory rather than on generative rules.

On average, each of the 181 polyvalent adjectives in my database has been found to conventionally occur in 5,3different valency patterns (including only predicative uses and where differing copulas are not counted as differentpatterns). Let us take the adjective redd ‘afraid’ as an example:

(39)

a. Eg er redd! (LNC) I am afraid! ‘I am afraid!’

F

b.

Eg er redd deg (LNC) I am afraid you ‘I am afraid of you’

ig. 2

c.

Jeg er redd for deg (NWC)

.

I

am afraid for you ‘I am afraid of you’

Net

d.

Eg er redd å vera åleine (LNC)

wor

I

am afraid to be alone ‘I am afraid to be alone’

k

e.

du er redd at eg blir borte (LNC) you are afraid that I become gone ‘You are afraid that I will disappear’

of valency constructions.

T.A. Haugen / Lingua 140 (2014) 83--9998

In (39), we see that redd can occur without a second complement (39a), with an NP object in (39b), with a prepositionalcomplement with the preposition for ‘for’ (39c), with an infinitive (39d), and with a finite clause complement (39e). Thereseems to be no other adjective in Norwegian that occurs in exactly the same patterns; hence, this is item-specificknowledge. One way of modelling this is to posit a network of valency constructions where the individual patterns arerepresented as nodes that are connected through the common element, viz. the predicator redd, see Fig. 2.

Valency constructions are equivalent to what Croft (2003, 2012) calls verb-specific constructions and to Boas’ (2003,2008, 2011) mini-constructions. It should be noted that valency as item-specific knowledge might also be represented in alexical model of valency; as argued by Croft (2003, 2012) and Welke (2009), the differences between a constructionalmodel and a lexical model are not necessarily that great, but see Haugen (2013) for a discussion of lexical vs.constructional models of valency and where it is argued that adjectival valency does provide evidence in favour of aconstructional approach. In any case, however, the question of predictability in complement realisation is fundamental tohow a grammar should be modelled.

7. Conclusions

In this article two major approaches to complement realisation, the aspectual approach and the causal approach, havebeen assessed against data from adjectival valency in Norwegian. As is the case for most approaches to complementrealisation, these have been developed on the basis of verbal valency. Adjectival valency also needs to be accounted for,however, and a first step towards such an account is to ask whether approaches to valency developed on the basis ofverbs are also applicable to adjectives. It has been argued that constructions with adjectival predicators have eventstructures in much the same way as constructions with verbal predicators. It has also been shown that especiallyaspectual but also causal notions are relevant when describing the event structures of adjectival predicators. InNorwegian, the primary copular verbs, være ‘be’ and bli ‘become’ complement each other aspectually in that the latter isthe inchoative counterpart of the former, which primarily occurs in constructions describing stative events. However,aspect and causality seem to be of very limited value when it comes to predicting complement realisation in the adjectivalpredicators studied. Of course, these results are not necessarily valid for languages other than Norwegian, but the data dodraw into doubt the general relevance of the notions of aspect and causality in predicting complement realisation.

The aspectual and the causal approaches to complement realisation are two of the most prominent versions of what Irefer to as the Predictability Thesis, i.e. the hypothesis that complement realisation is predictable from general rules basedon the semantic structure or the event structure of the valency carrier. The Predictability Thesis reflects a strong tendencyin the research on complement realisation, viz. that valency is an area of research which has been heavily theorised, butwhich has not been subject to empirical investigation to the same extent. In more empirically-oriented approaches, like thevalency grammars developed in the German grammatical tradition (cf. Heringer, 1996; Ágel, 2000; Engel, 2004; Welke,2011), the idiosyncratic nature of valency is emphasised. This is also reflected in the existence of valency dictionaries,which are found also for adjectives (Sommerfeldt and Schreiber, 1983). In this tradition, with its stronger empiricalorientation, no suggestion appears to have been made that valency is predictable. Herbst (2007:15) argues that, ‘‘[t]hephenomenon of valency is one part of the unpredictable, unsystematic aspects of language . . . it can hardly be considereda coincidence that valency research should have resulted in valency dictionaries since valency structures representidiosyncratic, word-specific types of information’’. The present corpus-based investigation of 181 polyvalent adjectives inNorwegian points in the same direction. Two central versions of the Predictability Thesis, the aspectual approach and thecausal approach, do not seem very promising when it comes to predicting complement realisation in adjectivalpredicators. Rather, the data from adjectival valency in Norwegian would also seem to place valency among theidiosyncratic rather than general aspects of language; it is necessary to specify both the semantic structure and therealisation of complements in the lexicon, with all the consequences that this has for grammatical theory.

Corpora

LBC. The Corpus for Bokmål Lexicography. The Text Laboratory, Department of Linguistics and ScandinavianStudies, University of Oslo. URL: http://tekstlab.uio.no/glossa/html/index_dev.php?corpus=bokmal.

LNC. NO 2014 Nynorskkorpuset. Department of Linguistics and Scandinavian Studies, University of Oslo. URL: http://no2014.uio.no/korpuset/.

NWC. NoWaC (Norwegian Web as Corpus). The Text Laboratory, Department of Linguistics and ScandinavianStudies, University of Oslo. URL: http://www.tekstlab.uio.no/nowac/.

Acknowledgements

I would like to thank Hans-Olav Enger, Kristian Emil Kristoffersen, and the anonymous referees of Lingua for theirvaluable comments and suggestions. Naturally, however, the final product is my responsibility alone.

T.A. Haugen / Lingua 140 (2014) 83--99 99

References

Ágel, V., 2000. Valenztheorie. Gunter Narr, Tübingen.Arche, M.J., 2006. Individuals in Time: Tense, Aspect and the Individual/Stage Distinction. John Benjamins, Amsterdam.Baker, M.C., 1988. Incorporation: A Theory of Grammatical Function Changing. The University of Chicago Press, Chicago.Boas, H.C., 2003. A Constructional Approach to Resultatives. CSLI, Stanford.Boas, H.C., 2006. A frame-semantic approach to identifying syntactically relevant elements of meaning. In: Steiner, P.C., Boas, H.C., Schierholz,

S.J. (Eds.), Contrastive Studies and Valency: Studies in Honor of Hans Ulrich Boas. Peter Lang, Frankfurt, pp. 118--149.Boas, H.C., 2008. Determining the structure of lexical entries and grammatical constructions in construction grammar. Annual Review of Cognitive

Linguistics 6, 113--144.Boas, H.C., 2011. Coercion and leaking argument structures in construction grammar. Linguistics 49, 1271--1303.Borer, H., 2005. The Normal Course of Events, vol. 2. Oxford University Press, Oxford.Carlson, G.N., 1977. Reference to Kinds in English. University of Massachusetts, Amherst.Chomsky, N., 1986. Knowledge of Language: Its Nature, Origin, and Use. Praeger, New York.Croft, W., 1991. Syntactic Categories and Grammatical Relations. University of Chicago Press, Chicago.Croft, W., 1998. Event structure in argument linking. In: Butt, M., Geuder, W. (Eds.), The Projection of Arguments: Lexical and Compositional

Factors. CSLI Publications, Stanford, pp. 21--63.Croft, W., 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford University Press, Oxford.Croft, W., 2003. Lexical rules vs. constructions: a false dichotomy. In: Cuyckens, H., Berg, T., Dirven, R., Panther, K.-U. (Eds.), Motivation in

Language: Studies in Honour of Günther Radden. John Benjamins, Amsterdam, pp. 49--68.Croft, W., 2012. Verbs: Aspect and Causal Structure. Oxford University Press, New York.Daugaard, J., 2002. On the Valency of Danish AdjectivesCopenhagen. http://bach.arts.kuleuven.be/PA/papers/daugaard/Dowty, D., 1991. Thematic proto-roles and argument selection. Language 67, 547--619.Engel, U., 2004. Deutsche Grammatik. Iudicium, München.Faarlund, J.T., Lie, S., Vannebo, K.I., 1997. Norsk referansegrammatikk. Universitetsforlaget, Oslo.Faulhaber, S., 2011. Verb Valency Patterns: A Challenge for Semantics-based Accounts. De Gruyter, Berlin.Frawley, W., 1992. Linguistic Semantics. Lawrence Erlbaum, Hillsdale, NJ.Goldberg, A., 1995. Constructions: A Construction Grammar Approach to Argument Structure. University of Chicago Press, Chicago.Goldberg, A., 2006. Constructions at Work: The Nature of Generalization in Language. Oxford University Press, Oxford.Haugen, T.A., 2012. Polyvalent Adjectives in Norwegian: Aspects of Their Semantics and Complementation Patterns. (Dissertation) Department

of Linguistics and Scandinavian Studies, University of Oslo.Haugen, T.A., 2013. Adjectival valency as valency constructions: evidence from Norwegian. Constructions and Frames 5, 35--68.Heltoft, L., 1995. Danish predicative adjectives and adverbials as valency bearers. In: Schösler, L., Talbot, M. (Eds.), Studies in Valency I: Nouns,

Adjectives, Verbs. Odense University Press, Odense, pp. 211--235.Herbst, T., 1983. Untersuchungen zur Valenz englischer Adjektive und ihrer Nominalisierungen. Gunter Narr Verlag, Tübingen.Herbst, T., 2007. Valency complements or valency patterns? In: Herbst, T., Götz-Votteler, K. (Eds.), Valency: Theoretical, Descriptive and

Cognitive Issues. De Gruyter, Berlin, pp. 15--36.Herbst, T., Schüller, S., 2008. Introduction to Syntactic Analysis: A Valency Approach. Narr Francke Attempto Verlag, Tübingen.Heringer, H., 1996. Deutsche Syntax Dependentiell. Stauffenburg, Tübingen.Hopper, P., Thompson, S.A., 1980. Transitivity in grammar and discourse. Language 56, 251--299.Jackendoff, R., 1996. The proper treatment of measuring out, telicity, and maybe even quantification in English. Natural Language and Linguistic

Theory 14, 305--354.Kennedy, C., Levin, B., 2001. Telicity corresponds to degree of change. In: 75th Annual Meeting of the Linguistic Society of America, Handout,

Washington, DC.Langacker, R.W., 1991. Foundations of Cognitive Grammar: Descriptive Application. Stanford University Press, Stanford.Levin, B., 1993. English Verb Classes and Alternations: A Preliminary Investigation. University of Chicago Press, Chicago.Levin, B., Hovav, M.R., 2005. Argument Realization. Cambridge University Press, Cambridge.Mourelatos, A.P., 1981. Events, processes and states. In: Tedeschi, P., Zaenen, A. (Eds.), Tense and Aspect. Academic Press, New York, pp.

191--212.Perlmutter, D., Postal, P., 1984. The 1-advancement exclusiveness law. In: Perlmutter, D., Rosen, C.G. (Eds.), Studies in Relational Grammar 2.

University of Chicago Press, Chicago, pp. 81--125.Pinker, S., 1989. Learnability and Cognition: The Acquisition of Argument Structure. MIT Press, Cambridge, MA.Rosen, S.T., 1999. The syntactic representation of linguistic events. Glot International 4, 3--11.Sommerfeldt, K.-E., Schreiber, H., 1983. Wörterbuch zur Valenz und Distribution deutscher Adjektive. Niemeyer, Tübingen.Talmy, L., 1976. Semantic causative types. In: Shibatani, M. (Ed.), Syntax and Semantics, vol. 6: The Grammar of Causative Constructions.

Academic Press, New York, pp. 43--116.Talmy, L., 1988. Force dynamics in language and cognition. Cognitive Science 12, 49--100.Tenny, C., 1994. Aspectual Roles and the Syntax--Semantics Interface. Kluwer Academic, Dordrecht.Vendler, Z., 1957. Verbs and times. The Philosophical Review 66, 143--160.Voorst, J.V., 1988. Event Structure. John Benjamins, Amsterdam.Welke, K., 2009. Valenztheorie und Konstruktionsgrammatik. Zeitschrift für germanistische Linguistik 37, 81--124.Welke, K., 2011. Valenzgrammatik des Deutschen. De Gruyter, Berlin.