Polyvalent adjectives: A challenge for theory-driven approaches to valency

55
2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006 Polyvalent adjectives: A challenge for theory-driven approaches to valency Tor Arne Haugen Ivar Aasen Institute for Language and Literature, Volda University College, Norway Abstract The article presents a corpus-based investigation of polyvalent adjectives in Norwegian, and the following basic theoretical questions are assessed in light of these data: Do predicators like verbs and adjectives take complements? If yes: Is complement realisation predictable on the basis of the semantic structure of the predicator, including its complements? Two of the main theory-driven approaches to valency, the Projectionist model and the Exoskeletal model, provide very different answers to these questions in that the former assumes that valency is rule-based, whereas the latter assumes that predicators can in principle be freely inserted into any syntactic frame. The data studied suggest that the answer lies somewhere in between these extreme positions: Predicators do take complements in the sense that specific complementation patterns are idiosyncratically connected to individual predicators, but predicators to not project a certain syntactic frame. Hence, even though predicators do not uniquely project a certain syntactic frame, their combinations with syntactic frames are to a large extent idiosyncratic. Keywords: Valency; Adjectives; Corpus-based; Projectionist model; Exoskeletal model 1. Introduction Predicators and complements (or their equivalents) are fundamental notions in most theories of grammar, and all theories of grammar need to account for the number and types of complements with which a predicator can co-occur. Most previous research on valency has the following characteristics: (i) it has mainly been concerned with verbs and (ii) the emphasis has been on theorising rather than on large-scale empirical investigations, although important work has been done in the area of valency dictionaries. In this article, the focus is extended to adjectival predicators, and the analysis is based on a large-scale corpus investigation of this predicator class. The valency patterns of predicators are fundamental to the formation of grammatical theories. More specifically, the manner in which the realisation of complements in surface structure is posited to come about has profound consequences for the overall architecture of

Transcript of Polyvalent adjectives: A challenge for theory-driven approaches to valency

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Polyvalent adjectives:

A challenge for theory-driven approaches to valency

Tor Arne Haugen Ivar Aasen Institute for Language and Literature, Volda University College, Norway

Abstract

The article presents a corpus-based investigation of polyvalent adjectives in Norwegian, and

the following basic theoretical questions are assessed in light of these data: Do predicators

like verbs and adjectives take complements? If yes: Is complement realisation predictable on

the basis of the semantic structure of the predicator, including its complements? Two of the

main theory-driven approaches to valency, the Projectionist model and the Exoskeletal

model, provide very different answers to these questions in that the former assumes that

valency is rule-based, whereas the latter assumes that predicators can in principle be freely

inserted into any syntactic frame. The data studied suggest that the answer lies somewhere

in between these extreme positions: Predicators do take complements in the sense that

specific complementation patterns are idiosyncratically connected to individual predicators,

but predicators to not project a certain syntactic frame. Hence, even though predicators do

not uniquely project a certain syntactic frame, their combinations with syntactic frames are to

a large extent idiosyncratic.

Keywords: Valency; Adjectives; Corpus-based; Projectionist model; Exoskeletal model

1. Introduction

Predicators and complements (or their equivalents) are fundamental notions in most theories

of grammar, and all theories of grammar need to account for the number and types of

complements with which a predicator can co-occur. Most previous research on valency has

the following characteristics: (i) it has mainly been concerned with verbs and (ii) the

emphasis has been on theorising rather than on large-scale empirical investigations,

although important work has been done in the area of valency dictionaries. In this article, the

focus is extended to adjectival predicators, and the analysis is based on a large-scale corpus

investigation of this predicator class.

The valency patterns of predicators are fundamental to the formation of grammatical

theories. More specifically, the manner in which the realisation of complements in surface

structure is posited to come about has profound consequences for the overall architecture of

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

a grammatical model. At present, there seem to be three major conceptions of valency in the

literature, and although it will necessarily imply that many details and technicalities need to

be set aside, it is useful to compare the core ideas behind these models. A first distinction to

be made is between theory-driven vs. data-driven approaches, although this distinction has

to be treated with great care. Unfortunately, there does not seem to be much communication

between these approaches, and the present article might be seen as an attempt to start

filling this gap.

Two of the three approaches that will be discussed are theory-driven models, and

they will be referred to as the Projectionist model and the Exoskeletal model, respectively. In

the Projectionist model, it is assumed that complement realisation is predictable and follows

from so-called linking rules, i.e. rules that account for how complements of certain types are

realised (e.g. Levin, 1993). This is referred to as the Predictability Thesis. A more recent

alternative, pursued by proponents of the Exoskeletal model (e.g. Borer, 2005; Marantz,

2013), is to treat valency patterns purely as a matter of general syntactic frames in which

predicators can be inserted. Proponents of this model argue that restrictions on the valency

patterns of individual predicators are extragrammatical; they follow from general knowledge

about the world and conventions. The differences between these models are striking: In the

Projectionist approach, it is assumed that valency is based on rules in the grammar. In the

Exoskeletal approach, on the other hand, valency is extragrammatical.

The theory-driven models can be contrasted to data-driven approaches, which tend to

view valency as one of the idiosyncratic aspects of language, emphasising the need for

storage of patterns in memory. Such approaches have existed for a long time in the valency

component of the dependency grammars in the German grammatical tradition following

Tesnière (1959, 1980), see for example Engel (2004), Ágel (2000), and Welke (2011).

Interestingly, it seems to have been taken for granted in this tradition that valency is

idiosyncratic. This is evident from the fact that valency dictionaries have been published,

both for verbs (Helbig and Schenkel, 1973; Schumacher et al., 2004) and for adjectives

(Sommerfeldt and Schreiber, 1983). There is also a comprehensive valency dictionary of

English covering both verbs and adjectives (Herbst et al., 2004), see also Herbst (2007), and

a valency dictionary of Danish adjectives (Daugaard, 2004). If valency is idiosyncratic, it

means that both the semantic structure and the realisation of complements need to be stored

with individual predicators. More recent approaches within construction grammar also tend to

be more data-driven (Boas and Sag, 2012; Croft, 2001; Goldberg, 1995, 2006).

A view of valency as idiosyncratic was also held in early versions of geneative

grammar (Chomsky, 1965), but this has later been abandoned in favour of the Projectionist

model and the Exoskeletal model.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

The different approaches to valency disagree on the following basic questions, of

which the first is the most fundamental:

(1)

1. Do predicators like verbs and adjectives take complements at all?

If yes,

2. Is complement realisation predictable from the semantic structure of the

predicator, including its complements?

Although these questions are certainly fundamental for grammatical theory, they are not

questions that can be decided on the basis of theorising; they can only be decided on the

basis of empirical investigations. The aim of this article is to assess the questions in (1)

against data from a corpus-based investigation of polyvalent adjectives in Norwegian.

We start with a presentation of the descriptive framework and the data in section 2,

before the basic questions in (1) are further discussed and the foundation for the assessment

of these questions is laid in section 3. We then move on to the empirical investigation of

complement realisation in the subject of the clause in section 4 and in the predicate of the

clause in section 5. In section 6, the empirical investigation is summed up, and combinations

of complement realisations in the subject and in the predicate are considered. Finally, the

consequences for grammatical theory are discussed in light of the empirical findings in

section 7.

2. Descriptive framework and data

A polyvalent adjective is understood as an adjective taking at least two complements

denoting participants in the event,1 in at least one of its valency patterns. Complements are

complementations that are specific to certain (groups of) adjectives (hence, the complements

are said to be valency-bound). Complements are frequently referred to as arguments, but in

a valency framework the term complement is preferred, since more emphasis is placed on

the fact that there is a relationship of dependency between a valency carrier and its

complement(s), and on the fact that complements can be different types of phrases and

clauses. Furthermore, complements are understood to be symbolic structures in that they

consist both of a meaning and a syntactic realisation;2 all the complements that are

1 The term “event” is used to refer to all aspectual types of events, also to stative events; cf. Croft

(2012:34).

2 The question whether there are expletive complements shall not concern us here.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

considered in this study have both a semantic structure and a syntactic realisation, and one

of the questions to be assessed is whether the latter is predictable from the former.

In terms of syntactic realisations, polyvalent adjectives in Norwegian occur with the

following types of complements: NPs, finite clauses, infinitive clauses, and prepositional

complements. The term nominal complement will be used to comprise the first three of these

types. In addition, the term predicator will be used to refer to a valency carrier per se, and it

is neutral as to whether this is an adjective or a verb.

A distinction needs to made between complements in the subject of a clause and

complements in the predicate of a clause. Following Herbst and Faulhaber (2009) we will use

the term subject complement unit (SCU) to refer to a unit which functions as the subject of

the clause and which is realised by a complement of the predicator, and the term predicate

complement unit (PCU) to refer to a unit which is a constituent in the predicate of the clause

and which is realised by a complement of the predicator. This descriptive framework is quite

neutral with respect to specific syntactic theories and thus suitable for the task at hand; all

models of grammar have to account for complementation patterns of this kind.

In contrast to English, where polyvalent adjectives overwhelmingly take prepositional

or clausal PCUs, Norwegian adjectives do take all kinds of nominal complements, including

NPs.3 Hence, adjectives take the same types of complements as do verbs:

(2) a. Naboene er lei bråket

Neighbours.DEF are tired noise.DEF

‘The neighbours are tired of the noise’

b. De er glad at alt gikk bra

They are glad that all went well

‘They are glad that everything went well’

c. Denne leiligheten er verdt å kjempe for

This appartment.DEF is worth to fight for

‘This appartment is worth fighting for’

We see that the adjective lei ‘tired, fed up’ takes an NP PCU in (2a), the adjective glad ‘glad’

takes a finite clause in (2b), whereas the adjective verdt ‘worth’ takes an infinitive clause in

(2c). The same kinds of complements are found as SCUs of polyvalent adjectives. For

reasons of space, the main concern here will be nominal complements. A comprehensive

3 This is a problem for grammatical models where adjectives are assumed not to take complements in

the generative sense, see for example Hale and Kayser (2002), but this shall not concern us here.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

investigation of valency needs to also take valency-bound prepositional complements into

account, but the full range of this complement class will not be analysed here. A more

comprehesive analysis of prepositional complements is found in Haugen (2012), on which

this article is partly based.

The material comes from a corpus investigation of 181 polyvalent adjectives in

Norwegian, where these adjectives were found to occur, on average, in 5,3 different valency

patterns. This includes patterns with both nominal and prepositional complements.

Furthermore, only predicative uses of the adjectives are included, and different copular verbs

are not counted as different patterns.4 The adjectives were selected by browsing all the

adjectives in an electronic glossary, which includes 5950 adjective lemmas. This means that

the group of polyvalent adjectives amounts to three per cent of the total number of lemmas,

and the conclusions to be drawn from this study are of course only valid for the group of

polyvalent adjectives.

In the first step of the corpus investigation, a lexicographical corpus of Norwegian

Bokmål (LBC) and of Norwegian Nynorsk (LNC) were investigated (each ca. 100 mill.

words), so that both written standards of Norwegian are represented. The two written

standards are codifications of the same language, however, and no attempt is made to treat

them separately. The searches were restricted to 1,000 random occurrences for each

lemma, and the resulting KWIC concordances were studied. In a second step, additional

searches have been carried out in cases where intuitively possible valency patterns have not

been found attested in the first step. In this step, the Web corpus NoWaC (NWC, ca. 700

mill. words) was applied, in addition to some Google searches (WWW). Only patterns with

attested examples were included in the database, and in most cases, the attested examples

are also the patterns that are intuitively acceptable. The exceptions to this will be discussed

as we proceed. The abbreviations used to refer to the corpora above are henceforth used to

mark authentic examples, whereas self-constructed examples are unmarked.

3. The basic questions

In theoretical approaches to valency there is no agreement as to the basic questions in (1).

As alredy noted, the Projectionist model and the Exoskeletal model provide very different

4 There are considerable differences in the possibilities for valency realisation between adjectives in

attributive and predicative function, and this is an interesting problem in its own right, see Haugen

(2013) for a discussion of this and Bolinger (1967) and Meltzer-Asscher (2012) for discussions of the

predicative/attributive distinction in general. The more restricted possibilities for adjectives in attributive

function to take complements of course also means that it is more challenging to predict complement

realisation with adjectives in predicative function.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

answers. This certainly reflects the fact that previous research, especially in the anglo-

american tradition, has frequently focused on theorising rather than on large-scale empirical

investigation, cf. the comprehensive survey in Levin and Hovav (2005). They argue that,

“[s]yntacticians are most often interested in sweeping generalizations concerning argument

realization” (Levin and Hovav, 2005:3). This is understandable in light of the fact that

grammatical theorising in its nature is a search for maximal generalisations.

In broad terms, the theory-driven approaches to complement realisation can be

divided into lexical approaches (the Projectionist model) and syntactical approaches (the

Exoskeletal model). In the former, generalisations are sought in the mapping between the

semantic structure of predicators and the realisation of their complements. A well-known

problem for such rule-based projectionist models is the fact that predicators frequently occur

in more than one valency pattern; hence, it is not the case that the choice of a certain

predicator determines the number and types of complements. This variant problem is

arguably the main reason for the theoretical move from lexical approaches to syntactical

approaches to complement realisation (cf. Borer, 2005; Marantz, 2013). In the latter,

generalisations concerning the relationship between predicators and their complement

realisations are largely abandoned, and, in principle, all predicators can be inserted into any

syntactic frame. As noted above, these two approaches are strikingly different.

The generalisations made by both models need to be tested empirically on as large

data sets as possible, and they should not be taken further than the data allow. Importantly,

the basic questions in (1) cannot be decided on the basis of theorising, and the data

available to assess them are utterances in which the different predicators actually occur and

speakers’ intuitions about the limits within which they can be used. Electronic corpora

represent rich possibilities for large-scale empirical investigations that can place the answers

to these questions on a more solid empirical ground. Before we get to the empirical

investigation, however, we need to clarify how the questions in (1) can be assessed.

The first and most fundamental question is whether predicators like verbs and

adjectives take complements at all. Only if the answer to this question is in the affirmative

does it make sense to ask the second question in (1), whether complement realisation is

predictable. The first point we need to clarify is therefore what it means for a predicator to

take a complement.

The basic empirical observation underlying the notion of valency is that predicators

like verbs and adjectives cannot occur in any syntactic frame. For example, the adjective lei

‘tired, fed up’ can occur both with an NP PCU (2a) and with a finite clause PCU (3a),

whereas glad ‘glad’ from (2b) can occur with a finite clause PCU, but it cannot occur with an

NP PCU (3b):

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

(3) a. Naboene er lei at vi bråker

Neighbours.DEF are tired that we make.noise

‘The neighbours are tired of our noise’

b. *De er glad det

They are glad it

Hence, valency features are needed to account for such restrictions, which are also known

as subcategorisation (Chomsky, 1965), cf. Helbig and Schenkel (1973) for an outline of the

parallels between valency and subcategorisation. An important point demonstrated by the

examples in (2) and (3) is also that possible restrictions on complement realisations need to

be stated in terms of phrase types and clause types (and possibly even in terms of specific

lexical items); to say that the adjectives lei and glad both take objects would not be

sufficiently specific to account for the restrictions observed. Faulhaber (2011:297) argues

that the use of functional labels like subject and object in models of valency phenomena,

“leads to an underestimation of the actual range of formal uses in which a verb can be found

and the possible range of irregularities to be identified in this area.” She goes on to state that,

“only by looking beyond elements such as noun phrase objects and instead comparing them

with the various patterns of present-day English including clausal and particle complements

can it be attempted to approach the question of the actual extent to which form is predictable

from meaning” (Faulhaber, 2011:297). Furthermore, also other languages than English and

other valency carriers than verbs need to be considered.

Proponents of the Exoskeletal model argue that predicators do not take complements

in the sense that, as far as the grammatical system is concerned, verbs and adjectives are

roots that can be inserted into any syntactic frame. There are different versions of this

approach (Borer, 2005; Hale and Keyser, 2002; Marantz, 2013; Ramchand, 2008), but here

we are mainly interested in the core idea, namely that predicators can in principle be inserted

into any syntactic frame, and that restrictions on complementation is to be accounted for in

terms of world knowledge associated with the predicator roots.

Unlike the question of predictability, to which we turn next, the question of whether or

not predicators take complements can be assessed solely on the basis of formal

complementation patterns, and an overview of the complementation patterns of the individual

adjectives included in this investigation is found in the appendix.

The second question in (1) concerns the predictability of complement realisation.

What a speaker minimally needs to know about a certain predicator is how many

complements it can take, what the complements mean, and how they are realised. Hence,

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

formal complementation patterns are an important basis for assessing this question as well,

but in addition we need semantic analyses of the predicators including their complements.

The idea that the realisation of complements might be predictable on the basis of

semantics can be traced back at least to Grimshaw (1979), who claimed that, “[w]hatever the

degree of predictability that may exist, it is to be found in the semantic, and not the syntactic,

domain” (p. 318), and it has played an important role for the research in this area (e.g.

Chomsky, 1986; Geuder and Butt, 1998; Grimshaw, 1990; Levin, 1993; Pinker, 1989;

Randall, 2010; Rosen, 1999).

There exist several different versions of the Predictability Thesis. Two prominent

recent approaches are the aspectual approach (see, e.g. Tenny, 1994; Levin and Hovav,

2005: 86ff) and the causal approach (Croft, 1991, 1998), see also Croft’s (2012) attempt to

reconcile the aspectual and the causal approach. In these approaches, complement

realisation is connected to aspectual and causal facets of events, respectively. These

approaches have been tested against data from adjectival valency in Haugen (2014a). It was

concluded there that these approaches do not seem very promising when it comes to

predicting complement realisation in adjectival predicators. In particular, it was shown that

complement realisation seems to be largely independent from the aspectual properties

associated with different copular verbs. Here, we will focus on the classical and perhaps

most basic facet of event structure, namely the participant roles of the complements.

The number and types of participants, i.e. the participant roles they fill, are central

features in all event representations. Events with different participants constitute different

event types, and regardless of theory and grammatical model, semantic descriptions of

participants are needed to distinguish between these types. There seems to be agreement in

the literature that there is no universally valid set of participant roles, and this is a problem

which is particularly pressing for the Projectionist model, since it needs to be assumed that

speakers are able to identify the semantic features of events that are relevant in forming a

rule-based system of complement realisation. Furthermore, for participant roles to have any

value in the formulation of linking rules, they cannot be predicator-specific. Obviously, if we

end up with predicator-specific rules, we end up with a representation where the semantic

structure of a complement is paired with its realisation, and no rules are needed anyway.

In light of the severe problems connected to defining participant roles and to applying

them in theories of complement realisation (see, e.g. Croft, 1991:155ff; Dowty, 1991:553ff;

Levin and Hovav, 2005:Ch. 2), a very relevant question to ask is how speakers are able to

aquire a rule-based system such as that proposed in the Projectionist model. Language

learners are certainly faced with the same problems as the linguist in this regard.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

As argued above, however, semantic descriptions of participants are needed, and

they are, in addition to formal complementation patterns, important in an assessment of the

predictability of complement realisation. To assess predictability, we need to posit participant

roles that are applicable across predicators, and the basic prediction of the Predictability

Thesis will be that these roles should also be realised in the same way across different

predictors. No claim is made that the participant roles explicated in Table 1 below are a

universally valid list for adjectival predicators; they have been explicated solely in an attempt

to assess the Predictability Thesis, and at the very least they serve as a starting point for

such an assessment.

There are few previous attempts at formulating participant roles for adjectives. Bennis

(2000) applies roles that have been proposed for verbs, like AGENT, THEME and

EXPERIENCER, whereas Meltzer-Asscher (2012) applies roles that are clusters of the

valued features CAUSE CHANGE and MENTAL STATE. These notions are certainly

relevant, but they are not sufficient to discern between the different types of participants that

are found with the whole range of polyvalent adjectives. Herbst (1983), which is a large-scale

investigation of adjectival valency in English, shows that participant roles are useful to

compare the semantic structure of participants also in events denoted by this class of

valency carriers, although other roles are needed for adjectives than for verbs. Herbst’s roles

have served as a point of departure for the roles that have been applied in the present study,

and only participant roles denoted by participants in the events, i.e. roles that are realised as

NP and/or P + NP have been explicated. Clausal complements are treated in terms of clause

types.

The role explications have the form of paraphrases containing the adjectival

predicator and the participants as variables. As far as possible, the paraphrases are

reductive paraphrases in the spirit of the Natural Semantic Metalanguage approach to

semantic explications (Goddard, 2008, 2011; Goddard and Wierzbicka, 2002). The reason

for this is that good definitions need to be simpler than what is to be defined, and circularity

needs to be avoided as much as possible. Independently of whether the primes applied in

the Natural Semantic Metalanguage approach are recognised as absolute semantic

primitives, they assure that a very limited range of simple terms is used.

The participant roles that have been found necessary to cover the complement types

in the material are summarised in Table 1. The names of the roles JUDGE, RELATOR,

TOPIC, and POTENTIAL INFLUENCE have been directly adopted from Herbst (1983). The

roles will be discussed as the analysis proceeds.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

ZERO: Someone or something, the ZERO, is Adj. Maybe ZERO does not think that ZERO is Adj. Someone X

thinks that ZERO is Adj.

English examples: The car is red, The journey is dangerous

EXPERIENCER: Someone, the EXPERIENCER, thinks, knows, feels, sees or hears someone or something.

English examples: He is afraid of the dog, She is sorry for him

EXPERIENCER AGENT: Someone, the EXPERIENCER AGENT, thinks, feels, sees or hears someone or

something. Because of this, the EXPERIENCER AGENT directs something against X.

English example: She is annoyed with him

RELATOR: Someone or something, RELATOR X, is in some way like someone or something else, RELATOR Y.

At the same time, RELATOR Y is, in the same way, like RELATOR X.

English examples: He is like his father, They are similar to their parents

COMPARED: Someone or something, COMPARED X, is, in some way, not like someone or something else,

COMPARED Y. At the same time, COMPARED Y is, in some other but similar way, not like COMPARED X.

English example: Your computer is superior to mine, Your car is inferor to my car

JUDGE: Someone, the JUDGE, thinks about someone or something X and thinks that this X is Adj to JUDGE.

English example: She is dear to me, Nothing is foreign to us

RECIPIENT: X directs something Y against the RECIPIENT. Maybe the RECIPIENT does not think that X is Adj,

and maybe the RECIPIENT does not know about Y. Maybe X does not think that X is Adj to the RECIPIENT.

Someone does think that X is Adj to the RECIPIENT.

English examples: We are thankful to him, He is angry with her

CAUSE: Someone or something X is Adj because of someone or something, the CAUSE. Maybe the CAUSE

does not know that X is Adj. Maybe the CAUSE directs something against X, or maybe someone else directs

something against X involving CAUSE.

English examples: He is afraid of the dog, They are tired of the noise

POTENTIAL INFLUENCE: Someone or something X is Adj in relation to someone or something, the

POTENTIAL INFLUENCE. The POTENTIAL INFLUENCE could do something to X or something involving the

POTENTIAL INFLUENCE could do something with X.

English example: They are rid of the problem

TOPIC: Someone or something X is Adj. When someone Y says that X is Adj, this is said in relation to or when it

comes to someone or something else, the TOPIC. X does not direct anything against TOPIC, and TOPIC does

not direct anything against X. Maybe the TOPIC does not feel, see, or hear anything.

English examples: She is certain about that, They are agreed on the plan

CHARACTERISED: Something X is Adj for someone or something, the CHARACTERISED. Maybe the

CHARACTERISED is not at the same time Adj of X. Maybe the CHARACTERISED does not think that the

CHARACTERISED is Adj. Someone Y thinks X is Adj for the CHARACTERISED.

English example: This situation is not good for Norway, The service is not good enough for a restaurant like this

VALUE: Someone or something X is worth VALUE. X can also be in some way that makes someone Y want to

do VALUE. X can also have some VALUE when Y wants X and has something else Z with the same VALUE.

Because of Z, Y can have X if someone who has X wants Z.

English example: The company is worth a fortune, The actor is worthy of the role

WISH: When someone X is Adj, X being Adj is because X wants WISH or wants to do or know something that

involves WISH.

English example: She is eager for success

CONTENT: Someone or something, the CONTENT, is inside or on someone or something X. There can be

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

much CONTENT inside or on X at some time, and there can be no CONTENT inside or on X at some other time.

English example: The bucket is full of water

CONSEQUENCE: Someone or something X is Adj, and because of this CONSEQUENCE happens to X.

English example: He is sentenced to three years in prison

Table 1. Overview of participant roles.

4. SCU realisation: NPs, finite clauses, or infinitive clauses?

When we now turn to the empirical investigation, we shall follow the following strategy: First,

the empirical data will be presented, and then we will assess the basic questions in (1),

namely whether predicators, in our case adjectives, do take complements at all, and

secondly, whether the realisation of complements is predictable from the semantic structure

of the participants in the events.

All the polyvalent adjectives in the material can take a SCU in the form of an NP,

whereas the other possible SCU realisations, finite clauses and infinitive clauses, are

possible for a limited range of adjectives only (when appropriate, finite clauses and infinitive

clauses will be referrred to collectively as processual complements). Among the adjectives in

the database, 41 per cent (75 out of 181) have been found attested also with processual

SCUs. With processual SCUs, these adjectives are used to ascribe the property the adjective

denotes to a process; in other words, these adjectives are used to evaluate processes. In the

next section, we shall explore whether there is a correlation between the participant roles of

NP SCUs and the possibility of a processual SCU for the same adjectives.

4.1. Adjectives taking processual SCUs

The participant roles ZERO, EXPERIENCER, EXPERIENCER AGENT, RELATOR, and

COMPARED (Table 1) are all realised as SCUs in the form of NPs, and if we classify the

adjectives according to the role of their NP SCUs, the first observation to be made is that

there are adjectives in all groups that also take processual SCUs. First, we have the

adjective uforståelig ‘incomprehensible’ taking a ZERO SCU in (4a), and a processual SCU

in (4b):5

(4) a. dette er meg uforståelig! (NWC)

this is me incomprehensible

‘This is incomprehensible to me’

5 Most authentic examples of this kind have extraposed subjects, and due to their structure, such

constructions are much easier to find in corpora that are not parsed. It seems always to be the case,

however, that these subjects are also intuitively acceptable in the canonical subject position.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

b. Det er helt uforståelig å kutte et tiltak mot en sårbar

It is completely incomprehensible to cut a measure against a vulnerable

og utsatt gruppe (NWC)

and exposed group

‘It is completely baffling to cut off a measure directed at a vulnerable and exposed

group’

The ZERO role is the most general of the semantic-role explications. The idea of a ZERO

role has been adopted from Langacker (2008), who adopts this term “for participants whose

role is conceptually minimal and nondistinctive. This is the neutral or baseline role of

participants that merely exist, occupy some location, or exhibit a static property” (Langacker,

2008:356).

In some cases, the SCU can be interpreted both as ZERO and as an

EXPERIENCER. The EXPERIENCER explication states that “someone, the EXPERIENCER,

thinks, knows, feels, sees or hears someone or something.” In (5a), the adjective

verdig ‘worthy’ takes a ZERO/EXPERIENCER SCU, whereas it takes a processual SCU in

(5b):

(5) a. Solskjær er verdig å bli ny reservelagstrener (NWC)

Solskjaer is worthy to become new reserve.team.coach

‘Solskjaer is worthy of becoming the new reserve-team coach’

b. Det er ikke verdig at mennesker som i en fase av livet trenger hjelp skal

It is not worthy that people who in a phase of life need help shall

være prisgitt de valg som det offentlige gjør for deg (NWC)

be at.the.mercy the choices that the government does for you

‘It is not worthy that people who need help in a phase of life shall live at the mercy

of the choices that the government makes for you’

A third example is the adjective lei ‘tired, sorry, sad’, which takes an EXPERIENCER SCU in

(6a), and a processual SCU in (6b):

(6) a. Vi er lei for at han forlater oss (NWC)

We are sorry for that he leaves us

‘We are sorry that he is leaving us’

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

b. At kvinnene vert isolerte og får depresjonar er sjølvsagt leitt (LNC)

That women.DEF become isolated and get depressions is of.course sad

‘That the women become isolated and depressed is of course sad’

Fourth, we have the adjective ergerlig ‘annoying’ taking an EXPERIENCER AGENT SCU in

(7a), and a processual SCU in (7b):

(7) a. Jeg ble som rimelig kan være ergerlig på ham og sa at mine

I got as reasonably may be annoyed on him and said that my

anbefalinger burde telle mer (NWC)

recommendations should count more

‘I was, as reasonably may be, annoyed at him and said that my recommendations

should count more’

b. det er ergerlig at programmet ikke ble til (NWC)

it is annoying that programme.DEF not became to

‘It is a pity that the programme was not realised’

There is a difference compared to the example in (6a) in that the SCU in (7a) directs

something against the PCU. This is evident from the fact that (7a) is a quite natural answer to

a question like Hva skjer? ‘What is happening?’ Hence, this construction resembles the

aspectual class of activity (Vendler, 1957). The example in (6a) is not a possible answer to

this question, and this difference is reflected in the EXPERIENCER AGENT explication in

Table 1.

Fifth, we have the adjective identisk ‘identical’, which takes a RELATOR NP SCU in

(8a) and a processual SCU in (8b):

(8) a. Forfatteren har vært identisk med mannen som hadde ideen til boka (NWC)

Author.DEF has been identical with man.DEF who had idea.DEF to book.DEF

‘The author is identical to the man who had the idea for the book’

b. Å skru av lyset til ære for jorden er symbolsk sett identisk med to

To turn off light.DEF to honour for earth.DEF is symbolically seen identical with å

stikke en strikkepinne i hjernen (NWC)

stick a knitting.needle into brain.DEF

‘Turning off the lights in honour of the earth is symbolically identical to sticking a

knitting needle into your brain’

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

The RELATOR role is found with adjectives that relate and compare two entities in a

symmetric, reciprocal relationship. The difference between the SCU and the PCU is in this

case merely perspectival, in that the SCU serves as a primary reference point, whereas the

PCU serves as a secondary reference point in the reciprocal relationship. In some cases, the

compared entities are in an asymmetric, non-reciprocal relationship, and in such cases we

have the role COMPARED:

(9) a. Apples teknologi og løsning er overlegen konkurrentene (NWC)

Apple’s technology and solution is superior competitors.def

‘Apple’s technology and solution is superior to the competitors’

b. Det er overlegent at det faktisk finnes et miljø (WWW)

It is superior that there actually exists a community

‘It is fantastic that there actually is a community’

We see that adjectives taking COMPARED NP SCUs (9a) can also take processual SCUs

(9b). The examples in (5)–(9) show that adjectives taking all the roles identified for NP SCUs

are represented among the adjectives that also take processual SCUs. On the other hand,

there are adjectives in all groups that do not allow processual SCUs. This is valid for full ‘full’

taking a ZERO SCU in (10a), redd ‘afraid’ taking an EXPERIENCER SCU in (10b), rasende

‘furious’ taking an EXPERIENCER AGENT SCU in (10c), nær ‘near’ taking a RELATOR SCU

in (10d), and underlegen taking a COMPARED SCU in (10e):

(10) a. *At de kjørte bil var fullt

That they drove car was full

b. *At det kom en stor hund var redd

That it came a big dog was afraid

c. *At festen ble så bråkete ble rasende

That party.DEF got so noisy became furious

d. *At fjellet ligger der er nært

That mountain.def lies there is close

e. *At de tapte var underlegent

That they lost was inferior

What the data presented above show is that there are no clear correlations between

semantic types of NP SCUs and the possibilities for these adjectives to occur with

processual SCUs. The question of course arises whether these adjectives have something

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

else in common that may make it possible to predict the different realisations of their SCUs.

We will return to this question in 4.3 below. What is also clear from the data presented thus

far is that only a subset of the investigated adjectives allow processual SCUs, and, in the

next section, we shall see that among the adjectives that do take processual SCUs, there is

also variation in which kinds of processual SCUs that are found with individual adjectives.

4.2. Adjectives taking that-clauses and infinitive clauses only

The most common for adjectives taking processual SCUs is that they allow that-clauses and

infinitive clauses only. This is the case for 46 of the 75 (61 per cent) adjectives taking

processual SCUs. The Norwegian equivalents to that-clauses are clauses with the

subordinating conjunction at ‘that’. Here are examples with the adjective ergerlig ‘annoying’:

(11) a. det er ergerlig at programmet ikke ble til (NWC)

it is annoying that programme.DEF not became to

‘It is a pity that the programme was not realised’

b. det er ergerlig å bruke tid og penger på dette (NWC)

It is annoying to spend time and money on this

‘It is a shame to spend time and money on this’

4.3. Adjectives taking that-clauses, interrogative clauses, and infinitive clauses

Out of the 75 adjectives taking processual SCUs, 14 have been found both with that-clauses

and with interrogative clauses, in addition to infinitive clauses. Interrogative clauses are of

two kinds; cf. Daugaard (2002:Ch.12), viz. hv– (Nynorsk kv–) clauses ‘wh–clauses’ and

clauses with the subordinating conjunction om ‘if, whether’ (Bokmål also hvorvidt ‘whether’).6

Below, the group taking all kinds of processual SCUs is exemplified by kritisk ‘critical’:

(12) a. det er kritisk at lyset er riktig (NWC)

it is critical that light.DEF is right

‘It is critical that the light is right’

6 As Daugaard (2002:226) points out, there are om-clauses that resemble subjects, but

that are to be analysed as conditional clauses. A test that can be used to distinguish between

these categories is that the phrase eller ei ‘or not’ can be added to interrogative om-clauses

which are subjects, but not to conditional clauses.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

b. Hva som skjer videre er kritisk for forholdet (WWW)

What that happens further is critical for relationship.DEF

‘What happens next is critical for the relationship’

c. det er kritisk om svanehalsen er steril (WWW)

it is critical whether swan.neck.DEF is sterile

‘It is critical that the swan’s neck be sterile’

d. det er kritisk å få kjørt rapporten (NWC)

it is critical to get run report.DEF

‘It is critical that the report is run’

4.4. Adjectives taking that-clauses only

There are also a few adjectives that have been found with that-clauses only, viz. medviten

‘conscious’, samtidig ‘simultaneous’, umedviten ‘unconscious’, and ulykkelig ‘unhappy,

unfortunate’:

(13) Det er ulykkelig at rollen til privat sektor er rammet inn av den ideologiske

It is unfortunate that role.DEF of private sector is framed in of the ideological

agenda (NWC)

agenda

‘It is unfortunate that the role of the private sector is framed by the ideological agenda’

4.5. Adjectives taking interrogative clauses and infinitive clauses only

With uviss ‘uncertain’, only interrogative clauses and infinitive clauses have been found, not

that-clauses:

(14) a. Hva som til slutt vil komme ut av dette er uvisst (LBC)

What which to end will come out of this is uncertain

‘What will ultimately come out of this is uncertain’

b. Om Hallingdal vil bli omfatta av støttetiltaka er uvisst (LNC)

Whether Hallingdal will be covered by support.measures.DEF is uncertain

‘Whether Hallingdal will be covered by the support measures is uncertain’

c. Samtidig er det jo helt uvisst å vite hvor i køen man er

Meanwhile is it certainly quite uncertain to know where in queue.DEF one is

på en netthandel (NWC)

on a web.shop

‘Meanwhile, it is uncertain where one is in the queue in a web shop’

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

An interesting question is whether this is due to the prefix -u ‘-in, -un’. As we saw in (13),

other adjectives with this prefix do occur with that-clauses, and it is shown in Daugaard

(2002) (for Danish) that there are no systematic valency variation in adjective pairs with and

without this prefix, see also Haugen (2013) for Norwegian.

4.6. Adjectives taking that-clauses and interrogative clauses only

Finally, with the adjectives kjent ‘known’, klar ‘clear’, and viss ‘certain’, all kinds of finite

clauses7 have been found, but not infinitive clauses:8

(15) a. At det måtte gjerast store endringar, var klart (LNC)

That it must be.made major changes, was clear

‘That major changes had to be made, was clear’

b. Kva for form samarbeidet får, er ikkje klart (LNC)

Which for form cooperation.DEF gets, is not clear

‘Which form the cooperation will have is not clear’

c. Om det skal kalles konsernmodell er ikke klart (LBC)

Whether it shall be.called corporate.model is not clear

‘Whether it will be called a corporate model is not clear’

4.7. Preliminary discussion

The restrictions observed in the preceding sections underscore the need for complement

realisations to be stated in terms of specific phrasal and clausal types. Among the 181

polyvalent adjectives in the material, 41 per cent (75 out of 181) allow processual SCUs, and

in this group there are restrictions as to whether different types of finite clauses and infinitive

clauses are allowed. Hence, it is not the case that all adjectives can be used in any syntactic

7 When it comes to interrogative clauses, there are restrictions regarding negation with these

predicators, see Haugen (2013).

8 There are also two other groups represented in the material. First, some adjectives have

only been found with SCUs in the form of infinitive clauses. These are jevnbyrdig ‘equal’,

jevngod ‘equally good’, lik ‘like’, ulik ‘unlike’, takksam ‘thankful’, and usammenlignbar

‘incomparable’. Intuitively, however, these adjectives belong to the group taking assertive

clauses and infinitive clauses. In addition, there is one adjective, uforståelig ‘unintelligible’,

which has only been found with assertive clauses, hv–clauses, and infinitive clauses.

Intuitively, however, an om-clause is also possible with this adjective.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

frame, as the basic idea behind the Exoskeletal model would suggest. NPs, finite clauses,

and infinitive clauses are grammatical notions, and they need to be accounted for by the

grammar. In other words, an exoskeletal grammar will not be able to distinguish between

grammatical and ungrammatical sentences.

On the other hand, it is not the case that an individual adjective always projects a

certain syntactic frame, which is the basic idea on which the Projectionist model is founded.

The answer seems to lie somewhere in between these extreme views: There are clear

restrictions on which valency patterns individual predicators can occur in, but it is common

for predicators to occur in a variety of different patterns. This will become even clearer when

we turn to the realisation of PCUs in the next section.

The semantic analysis of SCUs has shown no clear correlations between the

semantic structure of NP SCUs and the possibility of processual SCUs, but it might of course

be that the processual SCU realisations follow from some other feature of these events.

What the adjectives taking processual SCUs have in common is clearly that they are used to

evaluate processes, but adding this as a valency feature does not allow us to predict

anything. The reason for this is that there would be no other basis for this feature than the

realisations themselves. Adding the semantic feature PROCESS as a valency feature to

those adjectives that do take processual SCUs would be like saying that an adjective takes

an NP SCU because it has an NP valency feature. Again, this does not allow us to predict

realisation from any rule independent from the formal realisation itself. In addition, we have

seen that there are clear restrictions on which types of processual SCUs that are allowed

with individual adjectives. The valency patterns with which they are associated constitute

very important parts of the overall properties of predicators, and it does not really make

sense to say that these properties follow from something else than the patterns themselves.

After all, speakers are always exposed to predicators in specific patterns, not to predicators

in isolation.

What we are left with seems to be that predicators are conventionally associated with

certain types of complements, and that it is these conventional patterns that give rise to the

restrictions that are found. Hence, conventional patterns are crucial to distinguish between

different predicators as different event descriptions and to distinguish between grammatical

and ungrammatical sentences. We will return to this in section 6. Now, we will turn to

adjectives that take nominal PCUs.

5. Adjectives with nominal PCUs

Nominal PCUs include NPs, finite clauses, and infinitive clauses. Among the 181 adjetives in

the database, 87 (48 per cent) take one or more of these types, whereas 34 (19 per cent)

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

take NP PCUs. As we shall see, there are many idiosyncrasies as to which types of nominal

PCUs the individual adjectives take and to the order of adjective and PCU. The first

differentiation to be made is between PCUs preceding the adjective, and PCUs following it.

5.1. NP PCUs preceding the adjective

A first important generalisation to be made is that PCUs preceding the adjective are always

NPs, and the first complement type we shall be concerned with are PCUs that obligatorily

precede that adjective and that denote either JUDGE or RECIPIENT participants (all the role

explications are given in Table 1). These PCUs are frequently a personal pronoun. Here are

some examples with JUDGE:

(16) a. Platon er meg kjær, men sannheten er meg enda kjærere (NWC)

Plato is me dear, but truth.DEF is me even dearer

‘Plato is dear to me, but truth is even dearer’

b. Intet menneskelig skal være meg fremmed! (NWC)

Nothing human shall be me foreign!

‘Nothing human shall be foreign to me!’

The other adjectives taking a JUDGE PCU in the material are fjern ‘distant’, likegyldig

‘indifferent’, nær ‘near’, ubegripelig ‘incomprehensible’, and uforståelig ‘unintelligible’. In

addition, such PCUs can denote the role RECIPIENT:

(17) a. Hennes fastlege har vært henne behjelpelig (WWW)

Her GP has been her helpful

‘Her GP has been helpful to her’

b. vi vil alltid være ham takknemlig for det (NWC)

we will always be him thankful for that

‘We will always be grateful to him for that’

The main difference between the RECIPIENT role and the JUDGE role is that the JUDGE

itself conceptualises another participant as having the property denoted by the adjective,

whereas in the case of RECIPIENT, the conceptualisation might be done by someone else

only. With the adjective takknemlig in (17b), the preadjectival PCU ham ‘him’ fits the

RECIPIENT explication; this participant is not the conseptualiser of thankfulness. Other

adjectives taking RECIPIENTs preceding the adjective are nådig ‘merciful’, tro ‘faithful’, and

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

utro ‘unfaithful’. The RECIPIENT role is similar to the role TARGET suggested by Pesetsky

(1995: 56).

The prepositional complement with for 'for' (17b) denotes a role referred to as

CAUSE. The CAUSE role is characterised by the fact that there is a causal relation between

the CAUSE PCU and the state of the SCU in the construction. For example, for det ‘for that’

in (17b) refers to the direct reason for the feeling of thankfulness. Furthermore, the CAUSE

explication states that a CAUSE PCU might imply that the CAUSE is directed against the

SCU (X in the explication). The importance of distinguishing between RECIPIENT and

CAUSE is also demonstrated by an example such as Sue is angry with Bill about the party,

discussed by Pesetsky (1995: 63), which also demonstrates that a RECIPIENT (with Bill) can

co-occur with a CAUSE (about the party).9

As discussed in Haugen (2014a), adjectives taking CAUSE PCUs pose the same

problems to the causal approach to complement realisation as verbs of the kind Croft

(1991:251) calls reverse verbs, like hear, receive, suffer, and undergo, where the SCUs are

endpoints in the causal chain.

The adjective skyldig ‘owing, guilty’, which is the only adjectival predicator that can take

two nominal PCUs, is a special case with a RECIPIENT PCU. Both PCUs can follow the

adjective (18a), but they can also precede it (18b), although this is a rather restricted

possibility, see Haugen (2013). In (18b), this adjective arguably even takes four complements

(including the subject):

(18) a. Er han ikke skyldig meg et svar? (NWC)

Is he not owing me an answer?

‘Does he not owe me an answer?’

b. Jeg er ham stor takk skyldig for denne opplevelsen (NWC)

I am him big thanks owing for this experience.DEF

‘I am deeply indebted to him for this experience’

The second PCU of skyldig, et svar in (18a) and stor takk in (18b), expresses a role referred

to as TOPIC. From the explication of this role it is important to note that the X in this

explication does not direct anything against the TOPIC, and that the TOPIC never directs

anything against X; there is no causal relationship between TOPIC and X. Hence, the TOPIC

9 Pesetsky (1995) refers to the participant role of about the party as SUBJECT MATTER, but it is

important not to confuse this role with my role TOPIC, to be discussed below. In my view, there is

clearly a causal relationship here in that something having to do with the party is the reason for the

feeling of anger.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

role should not be confused with Pesetsky’s (1995: 57) SUBJECT MATTER, which is more

similar to our CAUSE role.10

JUDGE PCUs are generally realised in preadjectival position, and it might, therefore, be

asked whether it is possible to formulate a linking rule based on this role. However, we have

alredy seen that RECIPIENTs can precede the adjective, and there are also other participant

roles that can be preposed:

(19) a. Dommerne var ikke oppgaven voksen, freste trener Billy Walsh (NWC)

Referees.DEF were not assignment.DEF mature, raged trainer Billy Walsh

‘The referees were not mature for the assignment, raged trainer Billy Walsh’

b. Hun var seg selv lik (NWC)

She was her self like

‘She was like herself’

c. Porsche var Peugeoten overlegen (NWC)

Porsche.DEF was Peugot.DEF superior

‘The Porsche was superior to the Peugeot’

d. Dette er ikke en rettsstat verdig (NWC)

This is not a law.state worthy

‘This is not worthy of a state governed by the rule of law’

The adjective voksen ‘adult, mature’ in (19a) takes a TOPIC PCU. When someone says that

the referees were not voksen, this is said in relation to oppgaven ‘the assignment’. In

addition, dommerne ‘the referees’, X, do not direct anything against oppgaven, and

oppgaven does not direct anything against dommerne. Hence, the PCU fits the TOPIC

explication. The adjective lik ‘like’ in (19b) is used to compare two entities in a reciprocal

relationship, and these entities express the same participant roles, which are called

RELATOR. With lik, the second RELATOR can also, in some cases, e.g. when it is a

reflexive pronoun as in (19b), precede the adjective. Also PCUs with the role COMPARED

can, with the adjectives overlegen ‘superior’ and underlegen ‘inferior’, be realised as

preposed NPs (19c). The PCU en rettsstat ‘a state of law’ in (19d) realises a role which

shows some similarities to the RELATOR and the COMPARED roles (see also 4.1). This role

will be referred to as CHARACTERISED. The main difference from the two other roles is that

10 An important point that arises from this investigation is that the need for role distinctions is clearly

correlative with the amount of predicators that are included in the study. The more predicators that are

studied, the more subtle distinctions turn up. In my view, this is a problem in its own right for

approaches to complement realisation that rely on participant roles in the formulation of linking rules.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

a CHARACTERISED participant is not compared to another participant in the same

construction.

There remains one adjective whose PCU can also precede the adjective, viz. verdt

‘worth’:11

(20) a. De har liksom ikke vært bryet verdt på samme måte som før (LBC)

They have somehow not been bother.DEF worth on same way as before

‘They have somehow not been worth the bother in the same way as before’

b. Nye plasser å vise frem reklamen på er gull verdt (LBC)

New places to show forward commercial.DEF on is gold worth

‘New places to showcase commercials are worth gold’

The participant role of these PCUs is not quite compatible with any of our previous role

explications. In fact, verdt is one of the very few adjectives in the material that takes the role

referred to as VALUE. The closely related adjectives verdig ‘worthy’ and uverdig ‘unworthy’

can also take NP PCUs with this role (21a) as an alternative to a CHARACTERISED PCU

(21b):

(21) a. Eller vil velgerne vise seg denne tilliten verdig? (NWC)

Or will voters.DEF show themselves this trust.DEF worthy?

‘Or will the voters prove themselves to be worthy of this trust?’

b. Den nåværende situasjonen er ikke Norge verdig (NWC)

The present situation.DEF is not Norway worthy

‘The present situation is not worthy of Norway’

In (21a), the SCU is the CHARACTERISED,12 whereas the PCU expresses VALUE. In (21b),

on the other hand, the PCU Norge ‘Norway’ is the CHARACTERISED.

It is clear from the data presented above that although the position preceding the

adjective is restricted to a few adjectives taking NP PCUs, there is variation in the participant

roles that are realised in this position. Still, only JUDGE PCUs obligatorily precede the

11 This is a restricted possibility with verdt. Bryet ‘the bother’ in (20a) and gull ‘gold’ in (20b)

are fine in this position (the latter is, in fact, not acceptable in postadjectival position, but an

indication of value in terms of money is not: compare Bilen er verdt 100,000 ‘The car is worth

100,000 and *Bilen er 100,000 verdt ‘The car is 100,000 worth’), see Haugen (2013).

12 This means that we need to extend our list of SCU roles to include CHARACTERISED.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

adjective, and the JUDGE role is also overwhelmingly realised by a personal pronoun.

Hence, there is a linking generalisation to be made here: JUDGE PCUs are realised as

personal pronouns in preadjectival position, but this is also the only rule that can be

formulated on the basis of the suggested participant roles. As we shall see in the next

section, all the other roles that can be realised in preadjectival position can also be realised

in the position following the adjective.

A final point to note in relation to the Predictability Thesis is that some adjectives that are

semantically very similar to the adjectives taking preposed PCUs, do not occur in this pattern:

(22) ?Han var seg selv ulik

He was him self unlike

In (22), the adjective ulik ‘unlike’ is, according to my intuition, much less acceptable in this

pattern than its counterpart lik ‘like’ in (19b), even when the PCU is a reflexive pronoun. This

intuition is also supported by the corpora; there is no occurrence of the string [seg selv ulik]

(Nynorsk [seg sjølv ulik]) in any of the corpora.

The differences between participant roles are often subtle, and general roles are

difficult to explicate for adjectival predicators. As noted above, this is a problem in its own

right for attempts to formulate linking rules, and speakers are faced with the same problem.

However, the use of simple terms in the explications is helpful in the process of making such

subtle differences explicit.

5.2. NP PCUs with alternative positions

Some of the adjectives taking PCUs in preadjectival position can also take PCUs with the

same participant role in postadjectival position. This is the case with lik ‘like’, overlegen

‘superior’, underlegen ‘inferior’, verdig ‘worthy’, uverdig ‘unworthy’, and verdt ‘worth’:

(23) a. en nanometer er lik en tusendel mikrometer (LBC)

one nanometre is like one thousandth micrometre

‘One nanometre equals one-thousandth of a micrometre’

b. Xenon er overlegen halogen på de fleste områder (LBC)

Xenon is superior halogen on the most areas

‘Xenon is superior to halogen in most areas’

c. dette bordet er rett og slett ikke verdig en toppklasse-restaurant som Nodee

this table is right and simply not worthy a top.class-restaurant like Nodee

‘This table is simply not worthy of a top-class restaurant like Nodee’ (NWC)

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

d. Da han senere aksepterte embetet, fortalte Washington kongressen at

When he later accepted office.DEF, told Washington Congress.DEF that

han var uverdig æren (LBC)

he was unworthy honour.DEF

‘When he later accepted the office, Washington told Congress that he was unworthy

of the honour’

e. Denne vinen er verdt en test (LBC)

This wine.DEF is worth a test

‘This wine is worth a try’

We note that the variable position of the PCU is found with RELATOR (23a), COMPARED

(23b), CHARACTERISED (23c) and VALUE (23d)–(23e) participants. RELATOR,

COMPARED, and CHARACTERISED, especially, are similar roles; cf. the explications in the

overview of participant roles in Table 1, and two of the three adjectives taking the VALUE

role, verdig and uverdig, alternatively take the CHARACTERISED role. It might, therefore, be

argued that these roles form a coherent semantic group.

5.3. NP PCUs following the adjective

More problematic for the Predictability Thesis is the fact that there are also other adjectives

that take the RELATOR (24a) and the CHARACTERISED role (24c), which do not allow

variation in the position of the PCU. Hence, this is a predicator-specific feature:

(24) a. han var skyld far hennar (LNC)

he was related father her

‘He was related to her father’

b. *han var far hennar skyld

he was father her related

c. Jeg tar meg i å tenke at dette er typisk Norge (NWC)

I take myself in to think that this is typical Norway

‘I catch myself thinking that this is typical of Norway’

d. *Jeg tar meg i å tenke at dette er Norge typisk (NWC)

I take myself in to think that this is Norway typical

The RELATOR role is also found with skyldt ‘related’, nærskyldt ‘closely related’, and ulik

‘unlike’; the CHARACTERISED role is also found with utypisk ‘atypical’; and finally, the

COMPARED role is also found with overordna ‘superior’ and underordna ‘subordinate’.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Common to all these adjectives is that they have been found with postadjectival PCUs only.

Nær ‘near’ can also take a RELATOR PCU, and in such cases the PCU is always

postadjectival:

(25) a. mobiltelefonen er nær basestasjonen (LBC)

mobile.phone.DEF is near base.station.DEF

‘The mobile phone is near the base station’

b. *mobiltelefonen er basestasjonen nær

mobile.phone.DEF is base.station.DEF near

Also the following adjectives take postadjectival PCUs only: lei ‘tired’, and redd ‘afraid’, which

take the CAUSE role (26a), and kvitt ‘rid’, which takes the role POTENTIAL INFLUENCE

(26b):

(26) a. Folk er lei det kunstig høye markedet (NWC)

People are tired the artificially high market.DEF

‘People are tired of the artificially high market’

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

I got rid problem.DEF already after 17 hours

‘I got rid of the problem only after 17 hours’

The POTENTIAL INFLUENCE role is similar to TOPIC, but it differs from the latter role in that

a POTENTIAL INFLUENCE PCU potentially could do something with the SCU. In the

example in (26b), problemet ‘the problem’ might have influenced the SCU jeg ‘I’. A problem

represents a POTENTIAL INFLUENCE, although, in this case, it does not influence the SCU

at the time the utterance refers to. POTENTIAL INFLUENCE is also different from CAUSE in

that there is no causal relation between the NP bearing this role and the SCU having the

property denoted by the adjective.

As previously mentioned, an interesting difference between preadjectival and

postadjectival PCUs is that the former are restricted to realisation as NPs, whereas the latter

can also be realised as finite clauses and infinitive clauses (in addition to prepositional

complements). However, not all adjectives allow all the three possible nominal realisations. In

the following sections we will take a closer look at these restrictions.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

5.4. Adjectives taking NPs, that-clauses and infinitive clauses

First, we have adjectives taking all kinds of nominal PCUs. To allow also for a semantic

classification, we will classify them according to the role of their NP PCUs. All three

realisations are found with kvitt ‘rid’ taking a POTENTIAL INFLUENCE PCU; lik ‘like’ and ulik

‘unlike’ taking RELATOR PCUs; overordna ‘superior’ and underordna ‘subordinate’ taking

COMPARED PCUs; lei ‘tired’ and redd ‘afraid’ both taking CAUSE PCUs; and with verdig

‘worthy’ and verdt ‘worth’ both taking VALUE/CHARACTERISED PCUs. This is exemplified

below with lei ‘tired’:

(27) a. Svenskene er lei Microsoft (NWC)

Swedes.DEF are tired Microsoft

‘The Swedes are tired of Microsoft’

b. nå var jeg virkelig lei at alt skulle gå utover meg (NWC)

now was I really tired that everything would go out.over me

‘Now, I was really tired that everything would affect me’

c. jeg er ganske lei å høre om denne urettferdigheten (NWC)

I am pretty tired to hear about this injustice.DEF

‘I am pretty tired of hearing about this injustice’

5.5. Adjectives taking NPs only

The adjectives in this group are likestilt and likeverdig (both meaning ‘equal’), skyldt ‘related’,

nærskyldt ‘closely related’, overlegen ‘superior’, underlegen ‘inferior’, typisk ‘typical’, and

utypisk ‘atypical’. In this group, it is indeed possible to make some semantic generalisations.

First, many of the adjectives that are subject to this restriction take RELATOR- (likestilt,

likeverdig, skyldt, nærskyldt), COMPARED- (overlegen, and underlegen), and

CHARACTERISED PCUs (typisk and utypisk). As argued above, these roles are quite

similar, and these adjectives can be said to form a semantically coherent group. At first sight,

it therefore seems promising to formulate linking rules for these adjectives.

There is, however, a serious problem connected to our approach so far when it

comes to assessing the Predictability Thesis. When we give a general description of the

valency patterns of polyvalent adjectives, it is appropriate to group the adjectives based on

formal complementation patterns. This is also the point of departure for Levin’s (1993) study

of English verb classes. Crucially, however, in Levin's study, formal valency patterns are also

the basis for claiming that the adjectives are semantically similar. The problem with this,

when it comes to assessing the Predictability Thesis, is that it also needs to be considered

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

whether there are other predicators that are semantically similar and that do not occur in the

same patterns; cf. Faulhaber (2011:90). Hence, this approach might lead to biased

conclusions; the Predictability Thesis is very much a premise with this approach, rather than

a hypothesis. To really test this thesis, we need to also take other semantically similar

adjectives into consideration, and, indeed, there are other adjectives taking RELATOR-,

COMPARED-, and CHARACTERISED PCUs that are not subject to the NP-PCU-only

restriction. First, some of them allow all three realisation types, as already mentioned above.

In (28), this is exemplified with ulik ‘unlike’ taking a RELATOR PCU and with verdig ‘worthy’

taking a CHARACTERISED PCU:

(28) a. Robert var ikkje så ulik Joe (LNC)

Robert was not so unlike Joe

‘Robert was not so unlike Joe’

b. Dette er ikke ulikt at vi kan være 100 % sikker på at en stein har

This is not unlike that we can be 100 % certain on that a stone has

eksistert (WWW)

existed

‘This is not unlike the fact that we can be 100 per cent sure that a stone has existed’

c. Å bo i Spania er på mange måter ikke ulikt å bo hjemme på Frosta (NWC)

To live in Spain is on many ways not unlike to live home on Frosta

‘Living in Spain is in many ways not unlike living at home in Frosta’

d. Jeg synes ikke dette er verdig den politiske debatten i Norge (NWC)

I think not this is worthy the political debate.DEF in Norway

‘I do not think this is worthy of the political debate in Norway’

e. jeg er ikke verdig at du går under mitt tak (NWC)

I am not worthy that you go under my roof

‘I am not worthy of receiving you under my roof’

f. Spel og veddemål er ikkje verdige å vera rettshøve (LNC)

Games and betting are not worthy to be court.cases

‘Games and betting are not worthy of being tried in court’

Second, there are also adjectives taking RELATOR- and CHARACTERISED PCUs in the

form of predicator-specific prepositional complements, and to really assess the Predictability

Thesis, we need to consider the whole range of valency patterns. For reasons of space, we

will not treat the full range of prepositional complements here, but in (29) below are some

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

examples of adjectives taking RELATOR PCUs that can be realised as prepositional

complements only:

(29) a. Det gassiske språket er beslektet med indonesisk (NWC)

The Madagascan language.DEF is related to Indonesian

‘The Madagascan language is related to Indonesian’

b. *Det gassiske språket er beslektet indonesisk

The Madagascan language.DEF is related Indonesian

c. Jeg er kanskje ikke jevngod med spillerne i Rennes (NWC)

I am maybe not equally.good as players.DEF in Rennes

‘I might not be as good as the players in Rennes’

d. *Jeg er kanskje ikke jevngod spillerne i Rennes

I am maybe not equally.good players.DEF in Rennes

e. Overflaten skal være parallell med bunnen av filteret (NWC)

Surface.DEF shall be parallel to bottom.DEF of filter.DEF

‘The surface should be parallel to the bottom of the filter’

f. *Overflaten skal være parallell bunnen av filteret

Surface.DEF shall be parallel bottom.DEF of filter.DEF

In particular, the adjective beslekta ‘related’ in (29a) is semantically very close to the

adjectives skyldt ‘related’ and nærskyldt ‘closely related’, but does not share the same

complementation patterns. Jevngod ‘equally good’ in (29c) and parallell ‘parallel’ in (29e)

also fit the explication of the RELATOR role, but cannot take NP PCUs. The examples in (29)

demonstrate that when semantically similar adjectives that do not share the same valency

patterns are considered, the analysis yields less positive results for the Predictability Thesis

than would be the case with Levin’s (1993) methodology.

Prepositional complements with med ‘with’ is a common alternative realisation of

RELATOR PCUs, but it is not the only prepositional alternative, as demonstrated in (30):

(30) a. du er lik på mor di (LNC)

you are like on mother your

‘You are like your mother’

b. Norge er forskjellig fra Sverige (NWC)

Norway is different from Sweden

‘Norway is different from Sweden’

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

In (30a), we see that the adjective lik ‘like’ can take a prepositional complement with the

preposition på ‘on’ as an alternative realisation of the RELATOR role, and with forskjellig

‘different’ in (30b), RELATOR is realised by fra ‘from’.

We have discussed the adjectives taking NP PCUs only, but there are also other

restrictions at work among adjectives taking postposed nominal PCUs. When it comes to

finite clauses and infinitive clauses, the first thing to note is that finite-clause PCUs that are

not initiated by prepositions are always that-clauses.13 In most cases, at ‘that’ is obligatory;

the subordinating conjunction can only be omitted with glad ‘glad’ and redd ‘afraid’.

The possibilities of PCUs in the form of infinitive clauses and finite clauses is an area

where intuitions are sometimes uncertain and not fully in accordance with attested usage. An

investigation (Haugen, 2014b) of such cases in the Nordic Dialect Corpus (see Johannessen

et al. 2009) shows that this might be due to dialect differences, and such differences certainly

suggest that conventionality plays an important role in valency patterns. All attested cases

have been included, but cases that according to my own intuition are not clearly acceptable

will be marked “(?)” in the examples below.

5.6. Adjectives taking NPs and that-clauses only

The adjective var ‘aware, sensible’ is the only adjective that takes NPs and that-clauses, but

not infinitive clauses:

(31) a. Det var han som vart var henne (LNC)

It was he who became aware her

‘He was the one who became aware of her’

b. Nils vart var at Astrid heldt handa hans (LNC)

Nils became aware that Astrid held hand.DEF his

‘Nils became aware that Astrid was holding his hand’

c. *Nils vart var å halde handa hennar

Nils became aware to hold hand.DEF her

13 There is one possible exception to this, viz. the adjective var ‘aware’, which, according to

my own intuition, is possible also with a hv– ‘wh–’ clause: Han vart var kva som skjedde ‘He

became aware of what happened.’ Such an example has not been found in any of the

corpora, however.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

5.7. Adjectives taking NPs and infinitive clauses only

There are also adjectives that take both NPs and infinitive clauses, but not finite clauses.

First, there is a subtle difference between the counterparts verdig ‘worthy’ and uverdig

‘unworthy’. Only the former has been found with a finite clause; see (28e), here repeated as

(32a):

(32) a. jeg er ikke verdig at du går under mitt tak (NWC)

I am not worthy that you go under my roof

‘I am not worthy of receiving you under my roof’

b. ?Jeg er uverdig at du går under mitt tak

I am unworthy that you go under my roof

The constructed example in (32b) is perhaps acceptable, but such an example has not been

found in the corpora.14 Nær ‘near’ has also been found with an infinitive clause (33a), but not

with a finite clause (33b):15

(33) a. Ein fotgjengar var nær å bli køyrd ned sundag ettermiddag (LNC)

A pedestrian was close to be run down Sunday afternoon

‘A pedestrian was close to being run down on Sunday afternoon’

b. ?Ein fotgjengar var nær at han blei køyrd ned sundag ettermiddag

A pedestrian was close that he got run down Sunday afternoon

All the adjectives we have treated above have taken NPs as one possible PCU realisation.

We now turn to adjectives that are found with finite clauses and/or with infinitive clauses only.

5.8. Adjectives taking that-clauses and infinitive clauses only

We start with adjectives attested both with finite clauses and with infinitive clauses. The

clearly-acceptable cases in this group are heldig ‘lucky’, uheldig ‘unlucky’ (34a)–(34d) and

bevisst ‘conscious’ (34e)–(34f):

14 Another difference between these adjectives is that uverdig ‘unworthy’ has only been found with an

infintive clause PCU in a so called tough construction (Hicks, 2009; Langacker, 1995; Postal and

Ross, 1971): slike penger er uverdige å ta imot (WWW) ‘Such money is unworthy to receive’, whereas

this is not the case for verdig ‘worthy’, see (5a).

15 The adjective overordna ‘superior’ is also attested only with NP and infinitive PCUs, but according

to my intuition, an at-clause is also possible.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

(34) a. Jeg har vært heldig å få oppleve flyturer i svært mange forskjellige

I have been lucky to get.to experience flights in very many different

luftfartøy (NWC)

aircraft

‘I have been fortunate to experience flights in very many different aircraft’

b. Bestefaren hans var uheldig å kappa av fleire fingrar (LNC)

Grandfather.DEF his was unlucky to cut off several fingers

‘His grandfather was unlucky to have several fingers cut off’

c. Vi var heldige at vi begge hadde prøvd å trene veldig hardt (LBC)

We were lucky that we both had tried to train very hard

‘We were lucky that we had both tried to train very hard’

d. han er uheldig at en supermann helt utenom det vanlige er født på samme

he is unlucky that a superman wholly out.of the ordinary is born on same

tid (WWW)

time

‘He is unfortunate that a superman out of the ordinary was born at the same time’

e. Kirkerådet er bevisst at det kan være noen vansker med å

Church.Council.DEF is conscious that there can be some difficulties with to

innføre en slik ordning nå (NWC)

implement a such scheme now

‘The Church Council is aware that there may be some difficulties in implementing

such a scheme now’

f. Han uttaler at han ikke var bevisst å være kurder i Irak (LBC)

He says that he not was conscious to be Kurd in Iraq

‘He said that he was not conscious about being a Kurd in Iraq’

With glad ‘glad’, finite clauses are fully conventional, whereas infinitive clauses are attested,

but less acceptable according to my intuition:

(35) a. Jeg er glad at alt er i orden (NWC)

I am glad that everything is in order

‘I am glad that everything is in order’

b. (?)Er faktisk glad å slippe alle disse turistene hele tida (NWC)

Am actually glad to avoid all these tourists.DEF whole time.DEF

‘I am actually happy at being able to avoid all these tourists all the time’

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

The following additional adjectives are all attested both with finite clauses and with infinitive

clauses, but they are not clear cases: avhengig ‘dependent’; engstelig ‘anxious’; forbausa

and overraska (both meaning ‘surprised’); forberedt ‘prepared’; fornøyd and tilfreds (both

meaning ‘pleased’); skeptisk ‘sceptical’; stolt ‘proud’; and vant ‘accustomed’ (cf. Lødrup,

2004:83). Here are some examples:

(36) a. (?)Berg er engstelig at kjønnsforskningsmiljøene blir for små

Berg is anxious that gender.research.environments.DEF become too small

‘Berg is anxious that the gender-research environments become too small’

(NWC)

b. (?)jeg skal ha KS men er engstelig å ikke huske noe fra

I shall have Caesarean but am anxious to not remember anything from

dagen (WWW)

day.DEF

‘I will have a Caesarean, but am anxious not to remember anything from the

day’

c. (?)han er overrasket at det tok så lang tid (NWC)

he is surprised that it took so long time

‘He is surprised that it took so long’

d. (?)Du blir sikkert overrasket å høre fra meg (NWC)

You become surely surprised to hear from me

‘You will probably be surprised to hear from me’

The adjectives ergerlig ‘annoying’; sikker ‘certain’; sur ‘sour’; and takknemlig and takksam

(both meaning ‘thankful’) are attested but not clearly conventional with finite clauses, but they

are fully conventional with infinitive clauses in so-called tough constructions (Hicks, 2009;

Langacker, 1995; Postal and Ross, 1971). The following adjectives are found only with

infinitive clauses, and only in tough constructions: komfortabel ‘comfortable’, ukomfortabel

‘uncomfortable’, kritisk ‘critical’, positiv ‘positive’, trygg ‘safe’, utrygg ‘unsafe’, utakknemlig

‘ungrateful’, and uvant ‘unaccustomed’.

5.9. Adjectives taking that-clauses only

There are also adjectives that are attested only with finite clauses. It is true for all of these

that they are not fully conventional according to my own intuition (cf. Lødrup, 2004:83), but

the investigation in Haugen (2014b) suggests that they are fine in western and northern

varieties of Norwegian: bekymra ‘worried’, betenkt ‘thoughtful’, enig and samd ‘agreed’, uenig

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

and usamd ‘disagreed’, forbanna ‘angry’, fortvila ‘despondent’, irritert ‘irritated’, lykkelig

‘happy’, ulykkelig ‘unhappy’, misunnelig ‘envious’, misfornøyd ‘displeased’, nøgd ‘pleased’,

oppbrakt ‘exasperated’, opprørt ‘upset’, sjalu ‘jealous’, skuffa ‘disappointed’, urolig ‘worried’,

and viss ‘certain’. Here are two examples:

(37) a. (?)Nina ble litt bekymret at hun hadde vist for mye (NWC)

Nina became a.little worried that she had shown too much

‘Nina became a bit worried that she had shown too much’

b. (?)jeg var litt skuffet at ingen fra oss deltok (NWC)

I was a.little disappointed that no.one from us attended

‘I was a little disappointed that no one from among us attended’

5.10. Adjectives taking infinitive clauses only

Finally, there are also adjectives that are only found with infinitive clauses as PCUs and

where the infinitive clause is not part of a tough construction. These are behjelpelig ‘helpful’

(38a), which also occurs with a preposed NP; see (17a), and ferdig ‘finished’ (38b):16

(38) a. Vi er også behjelpelig å skaffe midlertidig husvære (NWC)

We are also helpful to find temporary accomodation

‘We can also help to provide temporary accommodation’

b. Redaktøren spør om eg snart er ferdig å skriva (LNC)

Editor.DEF asks if I soon am finished to write

‘The editor asks if I will soon be finished writing’

The survey of adjectives taking nominal PCUs is now complete. As was the case with SCUs,

there are a lot of idiosyncrasies when it comes to which nominal realisations are possible

with individual adjectives. Hence, the Predictability Thesis is faced with a lot of problems in

16 Lødrup (2004:83) mentions the following additional adjectives that can take only infinitive

clauses as nominal PCUs and that are polyvalent according to the definition applied here:

dyktig ‘clever’, flink ‘clever’, god ‘clever’ (god is fully conventional in tough constructions), ivrig

‘eager’, klar ‘ready’, and kompetent ‘competent’. With the exception of kompetent, they are all

attested with infinitive clauses, and are adjectives of the kind that, according to Faarlund et al.

(1997:406), are found with infinitive clause PCUs in western and northern varieties of

Norwegian, see also Haugen (2014b). In my own eastern variety, these adjectives are not

possible with infinitive clauses not initiated by prepositions.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

this area as well, especially when it comes to postposed nominal PCUs, but also when it

comes to the more restricted possibility of preposed NP PCUs.

6. Summary and discussion of results

As argued earlier, the two basic questions asked in (1) pertain both to formal

complementation patterns and to the relationship between semantics and formal realisation.

The question whether predicators do take complements at all can be decided solely on the

basis of formal complementation patterns, whereas the question of predictability cannot be

assesed on this basis alone, although formal complementation patterns certainly are one

factor in the assessment of predictability as well. In the summary below we will first take a

look at the main findings regarding formal complementation patterns, before we move on to

the relationship between participant roles and realisations.

6.1. Formal complementation patterns

In section 4 it was shown that 41 per cent (75 out of 181) of the investigated adjectives allow

processual SCUs. In addition, five different subclasses can be distinguished within the group

of adjectives taking processual subjects, so that we end up with the following classes:

Adjectives taking NPs only

Adjectives taking NPs, that-clauses and infinitive clauses

Adjectives taking NPs, that-clauses, interrogative clauses and infinitive clauses

Adectives taking NPs and that-clauses

Adjectives taking NPs, interrogative clauses and infinitive clauses

Adjectives taking NPs, that-clauses and interrogative clauses

Hence, it is clear that subjects are valency bound, in the sense that all adjectives cannot be

combined with the same types of SCUs. Meltzer-Asscher (2012), in contrast, argues that the

subjects of adjectives are not valency-bound, along the lines proposed by Marantz (1984),

Kratzer (1996), and others for verbs (although Meltzer-Asscher argues against such an

analysis in the case of verbs).17 Crucially, however, with the exception of so-called ergative

adjectives (Cinque 1990), to which we will return below, she seems to consider SCUs in the

form of NPs only, and it is not clear how different formal realisations such as those pointed

out above could be accounted for. The only processual SCUs considered by Meltzer-Asscher

17 This is a controversial issue within the generative framework. Horvath and Siloni (2008) argue that

so-called external arguments (i.e. subjects) are found as valency features of predicates in the lexicon.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

(2012) are SCUs of ergative adjectives like certain, which she claims always denote

properties of propositions and cannot take non-finite complements. The adjective certain,

however, can both take NP SCUs that do not denote propositions, as in he is certain he can

place it with a good publisher, and non-finite PCUs as in one side is certain to receive a

bloody nose, both from the British National Corpus. This demonstrates the importance of

considering the whole range of valency patterns when generalisations are made. In fact, the

Norwegian equivalent to certain, sikker, can take all types of SCUs.

When it comes to predictability from semantics, it is of course possible to stipulate

semantic features that would allow for exactly the patterns pointed out above, but it seems to

me that they would have no motivation independent from the complementation patterns that

they are supposed to predict. The same predicator in different complementation patterns

certainly describe semantically different events, but the most important clue to the

differences seems to be the different patterns themselves, which in their turn are posited to

be predicted by the semantic features. We will return to the circularity of such analyses

below.

Turning now to PCUs, three main types of nominal PCUs are found in Norwegian: NPs,

that-clauses, and infinitive clauses. Among the 181 adjetives in the database, 87 (48 per

cent) take one or more of these types, whereas 34 (19 per cent) take NP PCUs. PCUs

preceding the adjective are always NPs, whereas the following classes need to be

distinguished among adjectives taking postposed PCUs:

Adjectives taking NPs, that-clauses, and infinitive clauses

Adjectives taking NPs only

Adjectives taking NPs and that-clauses

Adjectives taking NPs and infinitive clauses

Adjectives taking that-clauses and infinitive clauses

Adjectives taking that-clauses only

Adjectives taking infinitive clauses only

Needless to say, PCUs are also clearly valency bound, and as argued earlier, the question

whether predicators take complements at all can be decided on the basis of formal

complementation patterns alone. It should be clear that categories such as NPs, that-

clauses, and infinitive clauses are grammatical categories and hence need to be accounted

for by the grammar.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

6.2 Formal complementation patterns and predictability

Formal complementation patterns can also tell us something about the second basic

question from (1), namely whether complementation is predictable from semantics. The

question is whether a rule-based system for the linking between semantic features and

formal complementation is plausible.

Léger (2010) proposes that adjectives taking processual PCUs in French fall into

three different classes, and that these classes are tied to the semantic features

PROPOSITION, EVENT, and ACTION, respectively. The first class allows for PCUs in the

form of infinitive clauses and tensed clauses in the indicative, the second class allows for

infinitive clauses and tensed clauses in the subjunctive, whereas the third class allows for

tenseless PCUs only. Again, it seems clear to me that these classes could not have been

established if it was not for the fact that they occur in excactly the complementation patterns

that they do. The analysis is circular in the sense that the formal complementation patterns

are the basis for positing the semantic features that are supposed to predict them. For

example, it is not possible to decide whether it is the semantic notion PROPOSITION that

leads to the realisation as a finite clause or whether it is the finite clause that leads to the

semantic notion of PROPOSITION. In fact, the alledged semantic classes are based on

formal complementation patterns just as much as they are based on semantic structure.

I agree with Langacker (2000:71) when he argues that, “the question of whether

grammar is predictable on the basis of meaning...is misguided, for it presupposes prior and

independent knowledge of what an expression means.” He argues further:

To ask whether meaning affords a basis for predicting grammar is consequently to

misconceive the basic logic of the situation. The problem is rather to formulate–

simultaneously–a characterization of semantic structure...as well as a characterization of

grammar that elucidates its role in the structuring and symbolization of conceptual content

(Langacker, 2000:71).

Based on valency differences between (near) synonymous predicators in English, Hudson et

al. (1996:442) make a similar point. They argue that, “it is hard to see how the alleged

meaning differences could be learned except on the basis of the syntactic differences that

they are meant to explain.”

Formal complementation patterns and the semantic structure of complements are two

sides of the same coin, since speakers are always exposed to both at the same time. It does

not follow from this, however, that there is a one-to-one relationship between meaning and

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

realisation, and that complementation can be plausibly predicted from semantics (or vice

versa) in a rule-based system. This is clear also from Léger’s (2010) analysis of adjectives

that do not fit into the generalisations made over the complementations of the adjectives in

the three classes with which she operates. For example, she proposes that adjectives such

as chanceux ‘lucky’ and malchanceux ‘unlucky’, which can take both tensed subjunctive

clauses and non-finite clauses, and adjectives such as brave ‘brave’ and gentil ‘nice’, which

only take infinitive clauses, take adjuncts rather than PCUs. It seems clear, however, that

speakers know excactly in which complementation patterns these adjectives in fact occur,

and that these complementation patterns are not acceptable for all adjectives. Hence, they

need to be treated as valency-bound complements, and when the whole range of valency

patterns in which a predicator occurs is considered, it turns out that these patterns to a large

extent are predicator-specific.

Consider the combinations of possible SCU realisations and possible PCU

realisations for the individual adjectives included in the present investigation. The set of

possible SCU realisations and the set of possible PCU realisations for a predicator can be

referred to as its complement inventory, and when the whole range of valency patterns is

taken into consideration, we end up with a large number of such inventories. The inventories

that are needed to account for the valency patterns of the adjectives in this investigation are

given in the appendix. It turns out that for the 181 adjectives investigated, 101 different

complement inventories are needed to account for the actually occurring realisations.18 This

means that for the different complement inventories there are, on average, only 1,8

adjectives that share the same possibilities.19 Furthermore, the analysis shows that 67 out of

the 181 adjectives have unique complement inventories, and the need for adjective-specific

rules renders a rule-based system implausible.

6.3. The relationship between participant roles and realisation

An attempt has also been made to analyse the relationship between participant roles and

complement realisations, and, as previously argued, a comprehensive assessment of

predictability needs to include also prepositional complements, since such complements

cannot be freely combined with any predicator either. Crucially, in the prepositional

18 In the complement inventories, prepositional complements are counted as one complement for each

valency-bound preposition.

19 The reservation has to be accepted, however, that the lists of adjectives in the different complement

inventories are not necessarily complete lists. On the other hand, adding additional adjectives to the

already-established inventories would not change the fact that all the inventories are needed to

account for the whole range of valency patterns.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

complements that were included in the investigation, the prepositions have specific meanings

in combination with certain adjectives; hence the combinations of adjectives and

prepositional complements are not open choice in Sinclair’s (1991, 2004) sense.20 The

prepositions that were considered are the following:

• av ‘of’

• etter ‘after’

• for ‘for’

• fra (Nynorsk: frå) ‘from’

• i ‘in’

• med ‘with’

• om ‘about’

• over ‘over’

• på ‘on’

• til ‘to’

Including prepositional complements, the possible realisations of the participant roles

explicated for PCUs in Table 1 are summed up in Table 2.

Participant role preA NP postA NP av etter for fra i med om over på til

RECIPIENT V V V V V

JUDGE V

CAUSE V V V V V V

POT. INFLUENCE V V V V V V

TOPIC V V V V V V V V V V V

RELATOR V V V V V V

COMPARED V V

WISH V V

CHARACTERISED V V V

CONTENT V V V V

CONSEQUENCE V

VALUE V V V V

Table 2. Participant roles and PCU realisations.

20 See Haugen (2012) for a more comprehensive discussion of this.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

To test the Predictability Thesis, it has been necessary to formulate participant roles that are

valid across predicators; predicator-specific roles would be of no use for the formulation of

linking rules. Another important point is that the search for semantic similarity has not been

based on formal complementation patterns only. This is crucial in order to avoid biased

conclusions about the Predictability Thesis; if only predicators that share the same valency

patterns are taken into consideration, the thesis is more of a premise than a hypothesis.

It is evident from the results in Table 2 that it is difficult to formulate rules for the

realisation of the PCU participant roles that have been explicated in this investigation. There

are two exceptions, viz. that valency-bound JUDGE participants are always realised as

preadjectival NPs, and that CONSEQUENCE participants are always realised by the P til.

Note, however, that a CONSEQUENCE participant is found with only one of the adjectives in

the material.

It is interesting to note that many of the prepositions realise a wide range of roles. The

preposition på even realises the same number of roles as do postadjectival NPs. The only

really predictable preposition from the perspective of realisation is om ‘about’, which realises

TOPIC only, but from the perspective of semantics, the TOPIC role can be realised by all

prepositions except fra ‘from’. What these data suggest is that conventional combinations of

adjectives and prepositions play a very important role in valency patterns, and that

complement realisations are probably stored with every single predicator.

It is common in models operating with linking rules that the rules apply between

participant roles and syntactic functions like subject and object, and that only NPs are

considered. This would, of course, simplify the model in some respects and decrease the

difficulties connected with formulating rules, but this approach would not capture the whole

range of valency phenomena that need to be accounted for.

Although semantically similar adjectives often occur in groups taking many of the same

valency patterns, it is also frequently the case that other semantically similar adjectives do

not share the same patterns. This finding is in line with Herbst’s (1983) investigation of

English adjectives; see also Faulhaber (2011:82). Herbst concludes that:

Die gegenseitige Zuordnung von semantisch definierten Adjektivgruppen und

syntaktischen Patterns hat in verschiedener Hinsicht deutliche Zusammenhänge

zwischen der Bedeutung von Adjektiven und ihrem Vorkommen in bestimmten

Satzmustern erkennen lassen. Eine 1 : 1 Entsprechung von Ergänzung (in einer

bestimmten semantischen Funktion) einerseits und semantisch definierter

Adjektivgruppe andererseits ließ sich allerdings nur in drei [von 35] Fällen nachweisen

(Herbst, 1983:245).

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

‘The mutual correspondence between semantically defined adjective groups and

syntactic patterns has shown clear correlations between the meaning of adjectives and

their occurrence in particular patterns. A 1 : 1 correspondence between complement

(in a specific semantic function) on the one hand and semantically defined adjective

groups on the other hand was, however, found only in three [out of 35] cases’.

In the present investigation, semantic groups of adjectives have not been separately defined,

but the importance of comparing semantically similar adjectives across groups based on

valency patterns has been emphasised, and the findings here point in the same direction as

Herbst’s investigation. The conclusion to be drawn from this is that conventional patterns

seem to play a very important role in valency phenomena, and that both the meaning and the

realisation of complements need to be represented in the complementation patterns of

individual predicators.

No claim can be made that the participant roles proposed for the complements of the

adjectives included in this investigation are necessarily the “correct” set for assessing

predictability.21 The same is true, however, for all approaches that rely on particpant roles in

the formulation of linking rules. As noted earlier, participant roles are notoriously difficult to

define, and, importantly, speakers are faced with the same problems. These difficulties are

per se a problem for theories involving linking rules based on participant roles, and at the

very least, the investigation has presented new data for the proponents of linking rules to

consider.

7. Consequences for models of grammar

All theories on valency need to account for data including the whole range of

complementation possibilities, and theories should be based on as large data sets as

possible. Theory-driven approaches to valency, however, like the Projectionist model and the

Exoskeletal model, are in general based only on a part of predicators’ complementation

possibilities; only coarse-grained constructions and alternations shared by bigger groups of

predicators are considered. The generalisations that have been found in this area are

certainly valuable, and Marantz (2013) is of course right when he argues that there are

general restrictions on complementation: “For example, no verb in English takes three

obligatory ‘internal arguments’ (ignoring the subject), and no verb may be followed by three

21 One of the reviewers of Lingua correctly points out that many of the role explications make subtle

distinctions. It should be noted, however, that if similar roles were lumped together, there would be

even more alternative realisations for each role.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

‘bare’ noun phrases (without prepositions)” (Marantz, 2013:152). Similar general restrictions

are valid also for adjectival predicators; there is for example no adjective in Norwegian that

takes more than two NP complements in the predicate, but this does not tell us very much

about the specific restrictions connected to individual predicators.

In light of large scale empirical investigations of valency, both for verbs (Boas, 2003;

Faulhaber, 2011; Helbig and Schenkel, 1973; Herbst et al., 2004; Schumacher et al., 2004)

and for adjectives (Daugaard, 2002, 2004; Herbst, 1983; Sommerfeldt and Schreiber, 1983;

and the present investigation), it certainly needs to be questioned whether the grammatical

model as a whole can be based only on constructions and alternations on a high level of

generality. There is a strong tendency in linguistics to equate “theoretical”, as in “theoretical

linguistics” with “maximal generality”. It needs to be acknowledged that storage in memory is

also a theoretical possibility.

One of the reviewers of Lingua raises the question whether there could be regular

rules that constrain but do not completely determine the realisation of participants. On the

specific level of valency, however, the data suggest that speakers need to have very specific

representations that enable them to use predicators correctly in the established patterns. Of

course, speakers are also able to analyse constructions into more schematic parts and use

them creatively, and it is clear that more general clause-type constructions are needed in

addition to specific valency constructions. However, also established usage needs to be

accounted for. Speakers do have intuitions as to what is and what is not an established

valency pattern for an individual predicator, i.e. they need to memorise very specific

representations. A general problem for models distinguishing between a general core and an

idiosyncratic periphery is also to account for how speakers are able to “decide” when to

“compute” on the basis of general rules and when to “retrieve” memorised patterns, and

memorised patterns are needed to be able to do generalisations in the first place.

Linguists, and perhaps also other speakers, may of course do higher-level

generalisations also in the area of valency; there are indeed groups of predicators that share

many of the same valency patterns, but speakers also know all the exceptions to more

general alternations. Thus, more schematic constructions can only be posited in addition to

specific valency constructions of the kind that have been explored in this investigation, and

the latter cannot be omitted from the model. As Müller and Wechsler (2014: 41) argue:

The notion of lexical valence structure immediately explains why the argument

realization patterns are strongly correlated with the particular lexical heads selecting

those arguments. It is not sufficient to have general lexical items without

valency information and let the syntax and world knowledge decide about argument

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

realizations, because not all realizational patterns are determined by the

meaning. The form of the preposition of a prepositional object is sometimes

loosely semantically motivated but in other cases arbitrary. For example, the

valence structure of the English verb depend captures the fact that it selects an

on-PP to express one of its semantic arguments... Such idiosyncratic lexical selection is

utterly pervasive in human language.

As already argued, the choice of the most plausible theory on complementation is an

empirical question, and the whole range of valency phenomena needs to be considered,

including other valency carriers than verbs. The investigation presented here is certainly also

a very restricted data set; only a group of Norwegian adjectives have been investigated, but

for these predicators, the whole range of valency patterns has been included. The

predicators that have been investigated occur in a variety of different patterns, which is in fact

one of the main reasons for the theoretical shift from lexical to syntactic approaches to

valency: “[D]espite ambitious attempts to describe how verbs might systematically appear in

a variety of syntactic structures depending on their semantic category (see in particular Levin,

1993), the flexibility of verbs to appear within the various set of frames relating form and

meaning has defied these efforts to regulate apparent alternations in argument structure

through the classification of verbs” (Marantz, 2013:152–153). The main insight of

constructional approaches to valency, both the Exoskeletal model and other constructional

approaches (e.g. Boas and Sag, 2012; Goldberg 1995, 2006), is that predicators do not

project a certain syntactic frame, and evidence from adjectival valency also suggests that

valency should be treated as a constructional rather than a lexical phenomenon (Haugen,

2013). The question is, however, on which level of specificity the constructions should be

assumed to operate.

The data from the present investigation suggest that valency patterns to a large

extent are predicator-specific and that it is predicator-specific constructions that are listemes

in the form of exemplar representations (cf. Bybee, 2010), not the predicator roots only. In

fact, the investigation has shown that a large number of predicators have unique complement

inventories. Hence, it is tempting to conclude with Lamb (2001), that “every lexeme has its

own syntax” (p. 177), and that “learning syntax is mainly a matter of learning lexemes” (p.

189). In our view, it is a matter of learning valency constructions, i.e. individual predicators

associated with a set of valency patterns. The term valency construction has been adopted

from Herbst and Schüller (2008), and is similar to Croft’s (2003, 2012) verb-specific

constructions and Boas’ (2003, 2008, 2011) mini-constructions. In valency constructions,

both the meaning and the realisation of complements are specified.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

When discussing exoskeletal models where there are no restrictions on syntactic

patterns connected to predicators, Ramchand (2008:3) argues that “[t]he actual limits on

variability reported in more standard accounts would then have to be due to limits based on

real-world knowledge and convention (extralinguistic).” The present investigation has shown

that it is highly problematic to account for the difference between grammatical and

ungrammatical sentences if conventions are excluded from grammatical knowledge. To make

such a step implies that the grammatical model cannot account for actually occurring

patterns. To account for the actually occurring valency patterns of the adjectives investigated,

the most plausible model of valency is a model which allows for a large repository of

conventionally occurring patterns. The model needs to to allow for a mental corpus in the

spirit of Taylor (2012). In such a model, conventionality is “the measure of well-formedness”

(Langacker, 1987:488).

8. Conclusion

In this article, an attempt has been made to assess two of the main theory-driven approaches

to complement realisation, namely the Projectionist model and the Exoskeletal model, on the

basis of a corpus investigation of 181 polyvalent adjectives in Norwegian. The investigation

includes the whole range of complementation patterns.

The Projectionist model and the Exoskeletal model are strikingly different in that the

former assumes that complementation is based on rules in the grammar, so-called linking

rules, whereas the latter assumes that valency is extragrammatical. Hence, these models

provide fundamentally different answers to the following basic questions:

(1)

1. Do predicators like verbs and adjectives take complements at all?

If yes,

2. Is complement realisation predictable from the semantic structure of the

predicator, including its complements?

The Projectionist model answers both of these questions in the affirmative. The Exoskeletal

model, in contrast, answers them in the negative.

The first of the questions in (1) can be answered on the basis of formal

complementation patterns alone, and the corpus investigation strongly suggests that both

complements in the subject and complements in the predicate are valency-bound, in the

sense that formal complement types like NPs, PPs, that-clauses and infinitive clauses can

only be combined with subsets of predicators. In fact, when combinations of subject

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

complements and predicate complements are considered, it turns out that 101 different

complement inventories are needed to account for the actually occurring realisations with the

181 adjectives studied. Hence, for the different complement inventories there are, on

average, only 1,8 adjectives that share the same possibilities. The analysis also shows that

67 out of the 181 adjectives have unique complement inventories, and the need for adjective-

specific rules renders a rule-based system implausible.

The predictability of complement realisations has also been assessed on the basis of

the participant roles of complements, and, even though participant roles need to be treated

with care, it seems clear that there is a one-to-many relationship between particpant roles

and their formal realisations, suggesting that the realisation of complements cannot be

predicted on the basis of their semantic structure.

The data from the present investigation thus suggest the following answers to our two

basic questions from (1): Even though predicators do not project a unique syntactic frame, as

the basic idea behind the Projectionist model would suggest, predicators do take

complements in the sense that they are idiosyncratically connected to a range of valency

patterns. Furthermore, it does not seem plausible that complement realisation is based on a

rule-based system for the linking of semantic structure to formal realisation, which means that

both the meaning and the realisation of complements need to be represented in the

complementation patterns.

Corpora

LBC. The Corpus for Bokmål Lexicography. The Text Laboratory, Department of Linguistics

and Scandinavian Studies, 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 Scandinavian Studies, University of Oslo. URL: http://www.tekstlab.uio.no/nowac/.

References

Ágel, V., 2000. Valenztheorie. Narr, Tübingen.

Bennis, H., 2000. Adjectives and argument structure. In Coopmans, P., Everaert, M.,

Grimshaw, J. (Eds.), Lexical Specification and Insertion. John Benjamins, Amsterdam,

pp. 27–67.

Boas, H. C., 2003. A Constructional Approach to Resultatives. CSLI Publications, Stanford.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Boas, H. C., 2008. Determining the structure of lexical entries and grammatical

constructions in construction grammar. Annual Review of Cognitive Linguistics 6, pp.

113–144.

Boas, H. C., 2011. Coercion and leaking argument structures in construction grammar.

Linguistics 49, pp. 1271–1303.

Boas, H. C., Sag, I. A., 2012. Sign-Based Construction Grammar. CSLI Publications,

Stanford.

Bolinger, D., 1967. Adjectives in English: Attribution and predication. Lingua 18, pp. 1–34.

Borer, H., 2005. Structuring Sense: The Normal Course of Events. Oxford University Press,

Oxford.

Bybee, J., 2010. Language, Usage and Cognition. Cambridge University Press, Cambridge.

Chomsky, N., 1965. Aspects of the Theory of Syntax. MIT Press, Cambridge,

Massachusetts.

Chomsky, N., 1986. Knowledge of Language: Its Nature, Origin, and Use. Praeger, New

York.

Cinque, G., 1990. Ergative adjectives and the lexicalist hypothesis. Natural Language and

Linguistic Theory 8, pp. 1–39.

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. Oxford University Press, Oxford.

Croft, W., 2003. Lexical rules vs. constructions: A false dichotomy. In Cuyckens, H., Berg,

T., Dirven, R., Panther, K. (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 Adjectives. Copenhagen. URL:

http://bach.arts.kuleuven.be/PA/papers/daugaard/ Accessed 2014-05-14.

Daugaard, J., 2004. Ordbog over danske adjektivers valens. Copenhagen. URL:

http://bach.arts.kuleuven.be/PA/papers/daugaard/ Accessed 2014-05-14.

Dowty, D., 1991. Thematic proto-roles and argument selection. Language 67, pp. 547–619.

Engel, U., 2004. Deutsche Grammatik. Iudicium, München.

Faarlund, J.T., Lie, S., Vannebo, K.I., 1997. Norsk referansegrammatikk. Universitetsforlaget,

Oslo.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Faulhaber, S., 2011. Verb Valency Patterns: A Challenge for Semantics-Based Accounts.

De Gruyter, Berlin.

Geuder, W., Butt M., 1998. Introduction. In Geuder, W., Butt, M. (Eds.), The Projection of

Arguments: Lexical and Compositional Factors. CSLI Publications, Stanford, California,

pp. 1–20.

Goddard, C., 2008. Cross-Linguistic Semantics. John Benjamins, Amsterdam.

Goddard, C., 2011. Semantic analysis: A Practical Introduction. Oxford University Press,

Oxford.

Goddard, C., Wierzbicka, A., 2002. Meaning and Universal Grammar: Theory and Empirical

Findings. John Benjamins, Amsterdam.

Goldberg, A. E., 1995. Constructions: A Construction Grammar Approach to Argument

Structure. University of Chicago Press, Chicago.

Goldberg, A. E., 2006. Constructions at Work: The Nature of Generalization in Language.

Oxford University Press, New York.

Grimshaw, J., 1979. Complement selection and the lexicon. Linguistic Inquiry 10, pp. 279–

326.

Grimshaw, J., 1990. Argument Structure. MIT Press, Cambridge Massachusetts.

Hale, K., Keyser, S. J., 2002. Prolegomenon to a Theory of Argument Structure. MIT Press,

Cambridge, Massachusetts.

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, pp. 35–68.

Haugen, T.A., 2014a. Adjectival predicators and approaches to complement realisation.

Lingua 140, pp. 83–99.

Haugen, T.A., 2014b. Adjektiv med utfyllingar: Geografisk og medial variasjon. In

Johannessen, J., Hagen, K. (Eds.), Språk i Norge og nabolanda: Ny forskning om

talespråk. Novus, Oslo, pp. 47–64.

Helbig, G., Schenkel, W., 1973. Wörterbuch zur Valenz und Distribution deutscher Verben.

Verlag Enzyklopädie, Leipzig.

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.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Herbst, T., Faulhaber, S., 2009. Optionen der Valenzbeschreibung: Ein Valenzmodell für das

Englische. In Konopka, M., Kubczak, J., Mair, C., Sticha, F., Waßner, U.H. (Eds.),

Grammatik und Korpora 2009: Dritte Internationale Konferenz, Mannheim, 22.-

24.9.2009. Narr Francke Attempo, Tübingen, pp. 411–428.

Herbst, T., Heath D., Roe, I. F., Götz, D., 2004. A Valency Dictionary of English: A Corpus-

Based Analysis of the Complementation Patterns of English Verbs, Nouns and

Adjectives. De Gruyter, Berlin.

Herbst, T., Schüller, S., 2008. Introduction to Syntactic Analysis: A Valency Approach. Narr

Francke Attempto Verlag, Tübingen.

Hicks, G., 2009. Tough-constructions and their derivation. Linguistic Inquiry 40, pp. 535–

566.

Horvath, J., Siloni, T., 2008. Active lexicon: Adjectival and verbal passives. In Armon-Lotem,

S., Rothstein, S., Danon, G. (Eds.), Generative approaches to Hebrew Linguistics.

John Benjamins, Amsterdam, pp. 105–134.

Hudson, R., Rosta, A., Holmes, J., Gisborne, N., 1996. Synonyms and syntax. Journal of

Linguistics, 32, pp. 439–446.

Johannessen, J.B., Priestly, J., Hagen, K., Åfarli, T.A., Vangsnes, Ø.A., 2009. The Nordic

Dialect Corpus: An advanced research tool. In Jokinen, K., Bick, E. (Eds.), Proceedings

of the 17th Nordic Conference of Computational Linguistics NODALIDA 2009. NEALT

Proceedings Series, Volume 4, pp. 73–80.

Kratzer, A., 1996. Severing the external argument from its verb. In Rooryck, J., Zaring, L.

(Eds.), Phrase Structure and the Lexicon. Kluwer, Dordrecht, pp. 109–137.

Lamb, S. M., 2001. Learning syntax: A neurocognitive approach. In Pütz, M., Niemeyer, S.,

Dirven, R. (Eds.), Theory and Language Acquisition. De Gruyter, Berlin, pp. 167–191.

Langacker, R. W., 1987. Foundations of Cognitive Grammar: Theoretical Prerequisites.

Stanford University Press, Stanford.

Langacker, R.W., 1995. Raising and transparency. Language 71, pp. 1–62.

Langacker, R.W., 2000. Grammar and Conceptualization. De Gruyter, Berlin.

Langacker, R. W., 2008. Cognitive Grammar: A Basic Introduction. Oxford University Press,

Oxford.

Léger, C., 2010. Sentential complementation of adjectives in French. In Cabredo Hofherr, P.,

Matushansky, O. (Eds.), Adjectives: Formal Analysis in Syntax and Semantics. John

Benjamins, Amsterdam, pp. 265–306.

Levin, B., 1993. English Verb Classes and Alternations: A Preliminary Investigation.

University of Chicago Press, Chicago.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Levin, B., Hovav, M. R., 2005. Argument Realization. Cambridge University Press,

Cambridge.

Lødrup, H., 2004. Clausal complementation in Norwegian. Nordic Journal of Linguistics 27,

pp. 61–95.

Maranz, A., 1984. On the Nature of Grammatical Relations. MIT Press, Cambridge MA.

Marantz, A., 2013. Verbal argument structure: Events and participants. Lingua 130, pp. 152–

168.

Meltzer-Asscher, A., 2012. The subject of adjectives: Syntactic position and semantic

interpretation. The Linguistic Review 29, pp. 149–189.

Müller, S., Wechsler, S., 2014. Lexical approaches to argument structure. Theoretical

Linguistics 40, pp. 1–76.

Pinker, S., 1989. Learnability and Cognition: The Acquisition of Argument Structure. MIT

Press, Cambridge, Massachusetts.

Pesetsky, D., 1995. Zero Syntax: Experiencers and Cascades. MIT Press, Cambridge,

Massachusetts.

Postal, P. M., Ross, J. R., 1971. !tough movement si, tough deletion no! Linguistic Inquiry 2,

pp. 544–546.

Ramchand, G., 2008. Verb Meaning and the Lexicon. Cambridge University Press,

Cambridge.

Randall, J. H., 2010. Linking: The Geometry of Argument Structure. Springer, Dordrecht.

Rosen, S. T., 1999. The syntactic representation of linguistic events. Glot International 4, pp.

3–11.

Schumacher, H., Kubczak, J., Schmidt, R., Ruiter, V., 2004. VALBU – Valenzwörterbuch

deutscher Verben. Gunter Narr Verlag, Tübingen.

Sinclair, J., 1991. Corpus, Concordance, Collocation. Oxford University Press, Oxford.

Sinclair, J., 2004. Trust the Text: Language, Corpus and Discourse. Routledge, London.

Sommerfeldt, K., Schreiber, H., 1983. Wörterbuch zur Valenz und Distribution deutscher

Adjektive. Niemeyer, Tübingen.

Taylor, J. R., 2012. The Mental Corpus: How Language is Represented in the Mind. Oxford

University Press, Oxford.

Tenny, C., 1994. Aspectual Roles and the Syntax-Semantics Interface. Kluwer Academic,

Dordrecht.

Tesnière, L., 1959. Éléments de syntaxe structurale. Paris.

Tesnière, L., 1980. Grundzüge der strukturalen Syntax. Klett-Cotta, Stuttgart.

Vendler, Z., 1957. Verbs and times. The Philosophical Review 66, pp. 143–160.

Welke, K., 2011. Valenzgrammatik des Deutschen. De Gruyter, Berlin.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

Appendix: SCU- and PCU realisations of individual adjectives

In the table below, the attested realisations of SCUs and PCUs of the individual predicators

are given. The table contains 181 adjectives in 101 different complement inventories. The

following abbreviations are used:

NP: noun phrase postposed to the adjective

NPpre: noun phrase preposed to the adjective

At: finite clause with the subordinating conjunction at ‘that’

Om: interrogative finite clause with the subordinate conjunction om ‘if, whether’

Wh: interrogative wh-clause

Inf: infinitive clause

Inftough: infinitive clause in tough construction

Valency-bound prepositional complements with the following prepositions have been

included:

av ‘of’

etter ‘after’

for ‘for’

fra (Nynorsk: frå) ‘from’

i ‘in’

med ‘with’

om ‘about’

over ‘over’

på ‘on’

til ‘to’

The complement inventories are given first, and the predicators that share the same

inventories are given on the following row.

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

SCU PCU

NP at om wh Inf NP NPpre at Inf Inftough av etter for fra i med om over på til

V V V V V

analog 'analogue'

V V V V

ansvarlig 'responsible', moden 'mature', umoden 'immature'

V V V V

arg 'angry', blid 'gentle', gretten 'grumpy', grinete 'cross', harm 'angry', rasende 'angry',

sint 'angry'

V V V V V V V V

avhengig 'dependent'

V V

avundsjuk 'envious', budd 'prepared', fattig 'poor', fysen 'yearning', huga 'desirous',

lysten 'lustful', merksam 'attentive', oppmerksom 'attentive', rik 'rich', spent 'excited',

uoppmerksom 'inattentive'

V V V V

behjelpelig 'helpful'

V V V V

bekymra 'worried', lykkelig 'happy', skuffa 'disappointed', urolig 'worried'

V V

beslekta 'related', forsiktig 'careful', fortrolig 'familiar', kombinerbar 'combinable',

kompatibel 'compatible', plaga 'tormented', ubeslekta 'unrelated', unøgd 'displeased',

varsom 'careful'

V V V

betenkt 'thoughtful', opprørt 'upset', oppbrakt 'exasperated'

V V V V V V V

bevisst 'conscious'

V V

blind 'blind', døv 'deaf', medansvarlig 'co-responsible', tom 'empty', uredd 'unafraid',

utakksam 'ungrateful'

V V V

byrg 'proud', kry 'proud'

V V V V

dyktig 'skilled', flink 'skilled', kompetent 'competent'

V V V V V

engstelig 'anxious'

V V V V V

enig 'agreed', samd 'agreed', uenig 'disagreed', usamd 'disagreed'

V V V V

entusiastisk 'enthusiastic'

V V V V V V V V

ergerlig 'annoying', sur 'sour'

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

SCU PCU

NP at om wh Inf NP NPpre at Inf Inftough av etter for fra i med om over på til

V V V

ferdig 'finished'

V V V V V

fjern 'distant'

V V V V V V V

forbanna 'angry'

V V V V

forbausa 'astonished', overraska 'surprised'

V V V V

forberedt 'prepared'

V V V V

forenlig 'consistent', sammenlignbar 'comparable', uforenlig 'incompatible'

V V V V V

fornøyd 'pleased', tilfreds 'content'

V V V V V V V

forskjellig 'different'

V V V V V

fortvila 'despondent'

V V V

forviten 'curious', nyfiken 'curious', nysgjerrig 'curious', sugen 'craving', sulten 'hungry'

V V V V

fremmed 'foreign', kjær 'dear', voksen 'adult'

V V V

fri 'free'

V V V V

full 'full'

V V V V V

gal 'crazy', vill 'wild'

V V V V V

gjerrig 'stingy', gnien 'stingy', identisk 'identical', knipen 'stingy'

V V V V V V

glad 'glad'

V V V V V V V V

god 'good'

V V V V V V V

heldig 'lucky', uheldig 'unlucky'

V V V V V

hissig 'short-tempered'

V V V V V

irritert 'irritated', sjalu 'jealous'

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

SCU PCU

NP at om wh Inf NP NPpre at Inf Inftough av etter for fra i med om over på til

V V V V

ivrig 'eager'

V V V V

jevnbyrdig 'equal', jevngod 'equally good', usammenlignbar 'incomparable'

V V V

jevn 'even', udyktig 'incompetent', ujevn 'uneven'

V V V V V V

kjent 'known'

V V V V V V V V

klar 'clear'

V V V V V

komfortabel 'comfortable', ukomfortabel 'uncomfortable', uvant 'unaccustomed'

V V V V V V V

kritisk 'critical'

V V V V

kvitt 'rid'

V V V V V V V V

lei 'tired'

V V V V V V V V

lik 'like'

V V V V V V V

likegyldig 'indifferent'

V V V V V

likestilt 'equal'

V V V V V V

likeverdig 'equal'

V V V V V V

medviten 'conscious'

V V

medskyldig 'complicit', uskyldig 'innocent'

V V V V

misfornøyd 'displeased', nøgd 'pleased'

V V V

misnøgd 'displeased', utilfreds 'discontented'

V V V V

misunnelig 'envious'

V V

nådig 'merciful', tro 'faithful', utro 'unfaithful'

V V V V V

nær 'near'

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

SCU PCU

NP at om wh Inf NP NPpre at Inf Inftough av etter for fra i med om over på til

V V V

nærskyldt 'closely related', skyldt 'related'

V V

nedstemt 'downcast', oppskaka 'upset'

V V V V

negativ 'negative', ukritisk 'uncritical'

V V V

oppglødd 'enthusiastic', ubekymra 'unconcerned', uroa 'troubled'

V V V V

opptatt 'occupied'

V V V V V

overlegen 'superior'

V V V V V

overordna 'superior'

V V V

parallell 'parallel'

V V V V V

positiv 'positive'

V V V V V

redd 'afraid'

V V V

samtidig 'simultaneous'

V V V V V V V V V

sikker 'certain'

V V V V

skeptisk 'sceptical'

V V V V V V V V V

skyldig 'guilty'

V V

stinn 'loaded'

V V V V V

stolt 'proud'

V V V V V V V

takknemlig 'thankful'

V V V V V V

takksam 'thankful'

V V V

trøtt 'tired'

V V V V V V

trygg 'safe', utrygg 'unsafe'

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

SCU PCU

NP at om wh Inf NP NPpre at Inf Inftough av etter for fra i med om over på til

V V

tvilende 'doubting'

V V V V V V

tydelig 'apparent', uklar 'unclear', usikker 'insecure', utydelig 'indistinct'

V V V V

typisk 'typical', utypisk 'atypical'

V V V V V V V

uavhengig 'independent'

V V V V V V

ubegripelig 'incomprehensible'

V V V V V

ubevisst 'unconscious'

V V V V V

uforståelig 'unintelligible'

V V V

uhuga 'reluctant'

V V V V V V

ukjent 'unknown'

V V V V V V V

ulik 'unlike'

V V V V V

ulykkelig 'unhappy'

V V V

umedviten 'unconscious'

V V V

underlegen 'inferior'

V V V V V V V V

underordna 'subordinate'

V V V V V

utakknemlig 'ungrateful'

V V V V V

V V V

uverdig 'unworthy'

V V V V V

uviss 'uncertain'

V V

uvitende 'unaware'

V V V V V

vant 'accustomed'

V V V V V

var 'aware'

2015. Lingua 156, pp. 70–100. http://dx.doi.org/10.1016/j.lingua.2014.12.006

SCU PCU

NP at om wh Inf NP NPpre at Inf Inftough av etter for fra i med om over på til

V V V V V V V

verdt 'worth'

V V V V V V V V V

verdig 'worthy'

V V V V V V V

viss 'certain'

V V V V V V V

åpen 'open'