TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.

1 (51)

A unified approach to semantic frames

and collocational patterns

Willy Martin

The aim of this chapter is to show that frames cannot only offer advantages in
explaining the meaning of words (as is usually claimed), but also in clarifying
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

their combinatorial behaviour. In order to do so, different kinds of frames are

presented, ranging from the more language-oriented ones à la Fillmore to the
more knowledge-in-general oriented ones à la Minsky to finally end up with what
I call ‘conceptual semantic frames’. The latter do not only serve as a background
for definitions but for word combinations and more particularly collocations as
well. In the frame-based approach presented here I try to show that the more
the collocator is, conceptually speaking, type-bound and the more it is, lexically
speaking, token-bound, the more we are dealing with a collocation that forms a
conceptual and lexical unit and therefore qualifies as a lexical collocation.

. Introduction

In a paper entitled ‘Double-decker definitions: The role of frames in meaning expla-

nations’, Fillmore (2003: 263) maintains that “certain kinds of structured background
information (or ‘frames’) should be treated as essential components or accomplish-
ments of word definitions”. Mutatis mutandis in what follows, I would like to show
that frames can offer advantages not only in explaining the meaning of words but also
in clarifying their combinatorial behaviour.

. Types of frames

There exist at least two ‘schools’ in frame linguistics, the one more language (i.e. syn-
tax) oriented (Fillmore and colleagues), the other more ‘knowledge’ (i.e. cognition)
Copyright 2008. John Benjamins Publishing Co.

oriented (e.g. Minsky).1 The Fillmorian school eventually led to the FrameNet project

. Frames as organisational devices are used not only in linguistics and cognitive studies but
also in political and economic studies. For the latter see, for example, Choices, Values and Frames,
ed. by D. Kahneman and A. Tversky, Cambridge/New York: Cambridge University Press, 2000.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 1/20/2021 6:33 AM via YONSEI UNIV
AN: 243182 ; Meunier, Fanny, Granger, Sylviane.; Phraseology : An Interdisciplinary Perspective
Account: s1096934.main.ehost
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.2 (52)

 Willy Martin

Table 1. Example of a ‘Fillmorian’ frame (based on Fillmore, Johnson & Petruck 2003)

SELL
ISA commercial transaction
SELLER who sells?
GOODS what is sold?
BUYER to whom is sold?
CONDITION for what is sold?

in which semantic frames of the sell type, for instance, were structured (see Table 1)
and defined as: “schematic representations of situations involving various participants,
props and other conceptual roles, each of which is a frame element. The semantic argu-
ments of a predicating word correspond to the frame elements of the frame or frames,
associated with that word” (from the FrameNet website2 ). Linked to the frame ele-
ments are their syntactic realizations (eg. PrepNP, NP etc. for BUYER). In the words
of the FrameNet makers: “FrameNet identifies and describes semantic frames, and an-
alyzes the meanings of words by directly appealing to the frames that underlie their
meanings and studying the syntactic properties of words by asking how their semantic
properties are given syntactic form” (Fillmore et al. 2003: 235).
Contrasting with Fillmore’s syntactico-semantic frame is the knowledge frame
that Minsky introduced in the mid-seventies in the domains of computational lin-
guistics and artificial intelligence. As the name suggests, Minsky’s aim was to develop
a model “that would enable to represent in a formalised way knowledge aspects linked
to human perception and the understanding of texts. This knowledge then would be-
come available for computers as well” (Wermuth 2005: 159–160; my translation from
Dutch into English). Although Fillmore subsequently moved from small-scale frames
(such as those used in Case Grammar) towards more large-scale cognitive ones as in-
troduced by Minsky, differences in orientation and depth between the two approaches
still remain (see Fillmore 2003: 288).
In my own search for an adequate lexical model I have tried to reconcile the more
general organisational Minskyan approach to knowledge with the more specific lexi-
cological aim of Fillmore. First of all I have taken over from Minsky the form, viz. the
slot-filler format where slots represent general conceptual relations and fillers the spec-
ifications of these general categories. As to the contents, Minsky sees frames as a kind
of expectation pattern that is evoked or adapted in a new situation. In Minsky’s own
words: “A frame is a data structure for representing a stereotyped situation (...). It is a
collection of questions to be asked about a hypothetical situation. It can be viewed as
an organised matrix of slots for given states of affairs” (Minsky 1977: 355). These states
of affairs can be quite diverse with Minsky. For instance, they can be entities such as
‘living room’ or events (scenes) such as ‘birthday party’. What the two entities have in

. FrameNet website: http://framenet.icsi.berkeley.edu/∼framenet

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.3 (53)

A unified approach to semantic frames and collocational patterns 

Table 2. Example of a ‘conceptual semantic’ frame (based on Martin 2003: 14)

MUSICAL INSTRUMENT
SLOTS FILLERS
ISA wind/percussion/stringed/plucked/keyboard . . . instrument
FUNCTION special function other than musical
MATERIAL
SIZE as compared to other members of the (sub)type
FORM
PARTS
MATERIAL
SIZE
FORM
NUMBER
ACCESSORIES
MATERIAL
SIZE
FORM
NUMBER
SOUND typical sound produced
USER typical user
MANNER OF PLAYING
POSITION OF PLAYER
POSITION OF INSTRUMENT vis-à-vis player
MECHANISM
ORIGIN
TYPICAL GENRE the instrument is used in
SIMILARITY similar to

common is that they can only be fully understood in a frame, a context, a background
of common knowledge based on cultural conventions.
Mainly inspired by Minsky, I have tried to represent the conceptual meaning of
lexical items by means of what I call ‘conceptual semantic frames’. To illustrate what I
mean, see the frame for musical instruments in Table 2. This shows the most important
features of these frames, viz.:
– Conceptual semantic frames are knowledge clusters that represent the conceptual
meaning of words in a slot-filler format; they do not make a strict distinction
between linguistic and encyclopaedic knowledge, but reflect the subjective, stereo-
typical knowledge of language users as evidenced in their usage (both passive and
active) of language.
– Conceptual semantic frames are definition models: they define the meaning of
words (as far as possible) by means of relations (slots) with other words, i.e.
lexicalised concepts.

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.4 (54)

 Willy Martin

Table 3. Semantic frame for violin using Collins Cobuild English Dictionary for Advanced
Learners, 3rd edition (2001) as a knowledge source

VIOLIN
ISA musical instrument
MATERIAL wood
PARTS
string
NUMBER: four
ACCESSORIES bow
MANNER-OF-PLAYING hold violin under chin and move bow across strings

– Conceptual semantic frames are type- or category-bound. In this respect the frame
in Table 2 is an empty, underspecified frame, the fillers only containing clarifica-
tions or pick lists. A frame can only get its specific fillers when dealing with a
concrete token or member of the category/type. So, for instance, the frame for the
token violin can be specified as shown in Table 3.

Using conceptual semantic frames transforms the lexicon into one large relational
network, frame or web with each word having its own vertical (paradigmatic) and
horizontal (syntagmatic) links. In the next section I will deal with these horizontal, i.e.
combinatorial, links in particular.

. Frames and the derivation of collocations

. Frames and collocational patterns

The advantages of a frame-based approach to the lexicon lie not only at the level of
representation (cf. Section 2), but also at the level of production. In particular, frames
can be extremely useful for producing collocations. As is well-known, constructing
a collocational dictionary or database is not an easy task. It does not suffice to take
a corpus and select the most frequent/statistically significant combinations from it.
Apparently there is no simple one-to-one relationship between frequency/statistics and
relevancy. Frames can be of help here in that they can evoke a collocational pattern.
When dealing with the frame for musical instruments, it becomes obvious that, as
with other artefacts, the function-slot takes a prominent place in defining the concept.
Corresponding to this definitional slot there is a collocational counterpart that can be
evoked. The following examples taken from German illustrate this. We start from the
generally accepted assumption that collocations consist of two parts: a base (the part
that is specified) and a collocator (the part that specifies):

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.5 (55)

A unified approach to semantic frames and collocational patterns 

Base Collocator
[MI = musical instrument] [F = typical functioning]
↓ ↓
the instantiation of a particular MI the lexicalization of a typical function-
ing of the particular MI
↓ ↓
der Gong ertönt/schlägt
(E. gong) (E. sounds)
From the slots given for musical instruments in Table 2, the following ‘collocations’
can now be derived:
– from the slot FUNCTION:
→ [MI] [typically functions]
e.g. der Gong ertönt/schlägt (the gong sounds)
→ [MI] does not function
e.g. der Gong schweigt (the gong is silent)
→ X [causes to function better] [MI]
e.g. eine Geige stimmen (to tune a violin)
→ [MI] [functions well]
e.g. die Geige hat einen guten Klang (the violin sounds pure)
→ [MI] [typically malfunctions]
e.g. die Geige klingt falsch (the violin sounds out of tune)
– from the slot SOUND:
→ [MI] [makes typical sound]
e.g. die Trompete schmettert (the trumpet blares)
– from the slot USER:
→ X [typically makes use of] [MI]
e.g. (auf der) Trompete blasen (to blow the trumpet)
As may be clear from the preceding discussion, the knowledge stored in a (type-bound)
frame not only helps us to understand utterances better, it also helps us to produce
utterances, in particular collocations. Of course, not all the combinations mentioned
above are considered to be collocations to the same degree, but here too frames can be
of help in differentiating between them.

. Frames and collocations

To understand which combinations are a subset of collocational patterns and there-

fore qualify as collocations, we need to define what collocations are. Although
Mel’čuk rightly points out that there is “no universally accepted formal definition of

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.6 (56)

 Willy Martin

collocations nor a proposal for their uniform and systematic treatment in dictiona-
ries” (Mel’čuk 1998: 23), most scholars do agree on two points. First, it is generally
accepted that collocations consist of two parts: a collocator and a base (see, inter
alia, Hausmann 2004 (quoting Schneider) using the term binemes as contrasted with
monemes). Second, it is usually agreed that these two elements show a degree of bind-
ing/fixation or restriction to each other, thus forming a unit that fits somewhere in
between idioms and free combinations. In this respect, some authors use the terms
‘semi-phrasemes’ (Mel’čuk 1998: 30) or ‘encoding idioms’ (Croft & Cruse 2004: 250),
implying that collocations, like idioms and phrasemes, behave as units.
As it is this latter feature in particular – the degree of binding/fixation/unification –
that remains vague and problematic, I will focus on it in some detail, starting from the
following working definition.
A collocation is
– a word group consisting of two conceptual elements: a collocator (determinans)
and a base (determinatum)
– of a certain syntactic type (N+N, V+N, A+N, Adv+V, Adv+A)
– showing a semantic, type-based, relationship between the two elements
– the more the collocator is, conceptually speaking, type-bound (bound to a con-
ceptual type or category) and the more it is, lexically speaking, token-bound
(bound to a lexical token/item), the more we are dealing with a collocation that
conceptually and lexically forms a unit, i.e. one that is a lexical collocation.

In the following I will comment on each part of the definition separately.

– A collocation is a word group consisting of two conceptual elements: a collocator
(determinans) and a base (determinatum)

By this I mean that a collocation is the combination of two concepts or frames which,
as a rule, are in a dependency relation, one (the collocator or modifier) modifying the
other (the base or head). So, for instance, in commit suicide we will consider suicide to
be the base or topic triggering the frame. This means that there is a dominant frame,
evoked by suicide and a dependent one (commit) zooming in on a specific aspect of the
dominant frame.3 Defining collocations as the combination of two concepts or frames
distinguishes them from idioms, which, although showing a compositional structure,
contain only one concept or frame.4
– of a certain syntactic type (N+N, V+N, A+N, Adv+V, Adv+A)

. Some authors, such as Poulsen 2005, prefer to speak of an ‘interdependent’ relationship,

even if there is a bias towards/dominance of one of the elements.
. Notice, however, that collocations can be recursively defined so that one collocation can
be embedded in another and thus lead to more than just two elements, as is the case in (take
((strong) (measures))).

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.7 (57)

A unified approach to semantic frames and collocational patterns 

This part of the definition implies that:

a. We call the combinations we are dealing with lexical in the sense that they com-
bine lexical items; combinations of a lexical item and a function word (preposi-
tion, conjunction etc.) are not taken into account here and will be regarded as
grammatical collocations.
b. The order in which the elements are given reflects a semantic dependency rela-
tionship, not a syntactical one: the left member is dependent on the right one. In
other words, the right member is considered to be the base (being the independent
or dominant frame), the left member the collocator (the dependent frame).
c. The dominant item here is the one that most strongly evokes the frame. Which
item is independent (dominant) and which is not is often a matter of dispute (see
e.g. Poulsen 2005: 271). In this discussion I take a functional point of view: in a
combination in which a noun and a verb occur, it is the noun, as a rule, that trig-
gers the frame functioning as the topic, while the verb fulfils the role of comment.5

– showing a semantic, type-based, relationship between the two elements

In a frame-based approach the combinatorial possibilities of lexical units (LUs) to

form constructions such as compounds and collocations are defined by the concep-
tual structure of the component LUs. In other words, given two LUs, X and Y, each with
their own conceptual frame, X and Y will only combine and unify if the modifying de-
pendent LU fits the slot of the modified dominant LU, thus specifying, among other
things, the meaning of the latter. In this respect there is no fundamental difference
in combinatorial behaviour between collocations and (a substantial subset of) com-
pounds, although their function may differ, collocations having a more characterising
and compounds a more categorising function.6
– the more the collocator is, conceptually speaking, type-bound (bound to a con-
ceptual type or category) and the more it is, lexically speaking, token-bound
(bound to a lexical token/item), the more we are dealing with a collocation that
conceptually and lexically forms a unit, i.e. a lexical collocation.

Although most definitions of collocations mention a degree of binding as a character-

istic feature, they usually remain rather vague in their elaboration of this phenomenon.
In fact, the more collocational a collocation is, the more its collocator is, conceptually

. In this respect it should be noted that the more concrete the meaning of the verb, the easier
it is to consider it the topic of the discourse. Compare for instance take in take measures (where
take is an abstract support verb) with take the car (take = make use of) and take a book (take =
grasp). In the last case (take a book) one can argue that take triggers the frame and book is one
of the many fillers of the slot/class of ‘graspable objects’.
. See, for example, Feilke 2004 who states, referring to Barz 1996, that ‘die Kollokation eher
charakterisiert, das Kompositium eher kategorisiert (z.B. dicke Milch vs. Dickmilch).’ (Feilke
2004: 54) [collocations rather characterise while compounds rather categorise].

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.8 (58)

 Willy Martin

speaking, type-bound and, lexically speaking, token-bound. This can only be under-
stood fully against the background of frames with their slot-filler format. In a concrete
combination such as drink coffee, for instance, drink acts as a filler for the slot ‘WAY OF
CONSUMING’, a slot that is typical for the type DRINK. In other words, for the mean-
ing of the word coffee, it is very relevant, just as it is for other drinks, that we drink it. On
the other hand, the word drink is not bound to the word coffee in an exclusive way. As a
matter of fact, drink is a default value for all drinks and so it is more type- than token-
bound. As a consequence, although there is a binding between the elements of drink
coffee, this binding is less strong than that between weak and coffee. This could suggest
that lexical combinatorics are subject to (type-bound) rules and exceptions. However,
linguistic reality is less straightforward. Besides rules and exceptions, the productiv-
ity/generality of rules also plays a role, as do preferences and prototypicality. I discuss
these phenomena below, illustrating them with Dutch examples (also see Martin 2006
and Martin 2007 for a more detailed discussion).

.. Type-bound regularity

Type-bound regularity in this context means that the lexicalisation of the collocator
is not an isolated phenomenon only applying to the base in question, but also char-
acterises the other members of the type the base belongs to. So, for instance, one may
suppose that the type ‘drink that has to be prepared’ has a ‘rule’ in Dutch that stipulates
that the slot ‘PREPARE’ is realized through the collocator (klaar) maken. As expected,
in Dutch we find words such as cocktail, soep (E. soup), bouillon (E. broth), gin-tonic,
grog etc. to which the rule applies, realizing the slot ‘PREPARE’ by means of the filler
maken.7 However, if we take koffie (E. coffee) or thee (E. tea) as possible tokens of the
type mentioned, we see that Dutch uses zetten here instead of maken. To explain this,
we can either take the stance that koffie and thee belong to a much more restricted type
than the one presented above, viz. something like ‘a drink that has to be prepared ac-
cording to a certain very specific procedure, i.e. by pouring hot water over leaves (tea)
or ground beans (coffee)’ and that therefore another rule applies, according to which
‘PREPARE’ is realized as zetten. Or we can adopt the point of view that rules must have
a sufficient degree of productivity/generality to be efficient/useful/acceptable. As this is
not true for koffie/thee zetten, we will regard the collocator as not being predictable by
a (more general) type-bound rule, but as an exception to such a rule. In other words,
zetten in the case of koffie/thee will be taken to be token- rather than type-bound.

.. Preferences
In the preceding discussion I suggested that the less productive the rules are that gov-
ern the lexicalization of collocators, the more these lexicalisations can be regarded
as token- rather than type-bound. There exists moreover another kind of token-

. In Dutch, as well as maken (E. make), we can also use mixen (E. mix) for cocktail and trekken
(E. make) for soep (E. soup) and bouillon (E. stock).

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.9 (59)

A unified approach to semantic frames and collocational patterns 

boundness. Consider the following case: in Dutch we have many words which belong
to the type ‘social allowance’. For instance: salaris (E. salary), uitkering (E. allowance),
pensioen (E. pension), kinderbijslag (E. child benefit), vakantietoeslag (E. holiday al-
lowance) etc. All these tokens have the slot ‘SIZE’ (the amount of the allowance) and
select a filler/collocator from a set of intensifiers such as groot, goed, hoog (for ‘high’)
and klein, laag (for ‘low’).8 However, if we take a closer look at these examples, we
see that there are differences in the preferred combinatorial behaviour of the mem-
bers. Salaris, for instance, prefers the company of goed, hoog and riant (for ‘high’), and
klein, laag, not slecht (for ‘low’); pensioen goes with goed and groot, less often with hoog
and riant (for ‘high’), and is only used with klein (for ‘low’), etc. What is clear here is
that, when different collocators can occur, their distribution over the various tokens
can lead to different predilections/preferences. In as far as this occurs, the combinato-
rial behaviour is only partially (if at all) type-bound and so expresses another form of
token-boundness.

.. Prototypicality
From this discussion a certain lexical categorisation in which rules, constraints and
preferences play a role, emerges. Another organising principle that we have to take
into account when dealing with collocations is that of prototypicality. Consider the
following concrete illustration: given a type ‘liquid to be consumed’, one expects a slot
‘(TYPICAL) WAY OF CONSUMING’. In Dutch this slot is filled by drinken (E. drink).
Drinken is thus a typical type-bound filler/collocator in this respect. Yet, depending on
the socio-cultural context, combinations as wijn (E. wine) drinken, wodka (E. vodka)
drinken, bier (E. beer) drinken, koffie (E. coffee) drinken, thee (E. tea) drinken etc. can all,
to a greater or lesser extent, also be considered to be token-bound. This is so because
the rule component of the lexicon need not just be taken to consist only of type-bound
abstract rules, but can also include concrete cases, and good prototypical/illustrative
token-bound examples for the rule in question. As one of my PhD students, Jeroen
Redel, has put it (personal communication): “this leads to the collocational paradox:
importing type-bound combinations into the more token-bound category via proto-
types”. In this sense koffie drinken, which is a typical type-bound collocation, can also
be considered token-bound, in as far as it acts as a prototypical example. In the same
vein, prototypical examples of free combinations can ‘intrude’ into semantic frames
and thus become type-bound. All in all, this leads to a typology and model such as
that presented and illustrated in Section 3.3 below.

. The choice of the collocator is, of course, also grammatically defined. Uncountables such as
kinderbijslag (E. child benefit) only take quantifiers such as veel (E. much) and weinig (E. little).

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.10 (60)

 Willy Martin

. Typology, model and illustration

.. Introduction
The typology for collocational patterns presented in Figure 1 emerges from the pre-
ceding discussion. Although this typology reflects the well-known idea of collocations
as ‘in-betweens’ situated on a continuum, it does so in a dynamic way: type-bound
collocations ‘intruding’ into the token-bound class via prototypes (e.g. koffie drinken
(E. drink coffee)) and free (non-bound) combinations doing the same with the type-
bound class (e.g. koffie inschenken (E. pour out coffee)), as shown in Figure 2.
In the next section I will take up the Dutch word koffie (E. coffee) to illustrate
the model. Of course, if the aim is to scale things up, a whole system of types, sub-
types, subsubtypes etc. is needed together with their frames (slots plus (default) fillers).
Therefore, I give below some heuristic clues which may serve as short cuts for rapidly
selecting token-bound lexical collocations from the set of potential ones. The following
clues can be used:
– The vaguer the collocators, the more difficult it is to predict (from the type) their
occurrence. Therefore, as a rule, light verbs (such as do a favour, take a look, have
breakfast etc.) will be token-bound collocators.

Collocational Patterns

Idioms Collocations Free Combinations

Grammatical Lexical Collocations

Collocations

Bound Free
(Token-Bound) (Type-Bound)

Figure 1. Types of collocational patterns

COLLOCATIONS COMBINATIONS

BOUND FREE FREE

(token-bound) (type-bound)

(intruding) prototypes (intruding) prototypes

Figure 2. Dynamic model for lexical collocations

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.11 (61)

A unified approach to semantic frames and collocational patterns 

– The more fillers there are for a certain slot (of a certain type), the more difficult it is
to predict their distribution and the preferences that play a role. As a consequence,
there is a high probability that, for example, intensifiers will be token-bound.
– The lexical functions as found in the work of Mel’čuk and his collaborators (see,
for instance, Mel’čuk 1998 and Mel’čuk & Zholkovsky 1988) and, in particular,
the syntagmatic standard lexical functions, have both a wide range of application
(they are general) and a strong impact on the meaning of the base-word. This
means that they typically yield type-bound fillers. However, the latter may result
in token-bound ones when they are strongly restricted and/or preferred.

In the following section I illustrate the model by taking all combinations of the word
koffie (E. coffee) as they occur in a well-known database for Dutch, the Referentie Be-
stand Nederlands (RBN) (see Martin & Maks 2005), characterising them according to
the categories mentioned in my typology/model. Figure 3 presents the combinations
in diagram format.

.. Examples and characterisation

Idioms and grammatical collocations
One of the data categories to be found in the RBN is called ‘combinatorics’. In fact, this
section covers more than just lexical collocations. It also includes idioms, pragmatic
formulae, proverbs, clichés, multi-word terms and the like, thus bringing this section
closer to phraseologisms in the broad sense, for which Schmitt and Carter (2004) use
the term ‘formulaic sequences’. I will start by examining two examples that occur in the
combinatorics section but, stricto sensu, fall outside my central object of investigation.
– dat is geen zuivere koffie (E. this looks suspicious; lit. this is not pure coffee).

What we have here is not a collocation, but an idiom operating as a single frame. It
could be argued that this is a figurative expression in which two literal concepts/frames
(that of coffee and that of flavour) are still visible. Figurative expressions thus bridge the
gap between idioms in the strict sense and collocations.
– op de koffie komen (E. to be invited for coffee).

Koffie/coffee is a polysemous word showing such meaning aspects as: the drink, the
product (beans/seeds) the drink is made of, the time when the drink is usually drunk,
the container the drink is served in, etc. In its temporal meaning koffie is typically used
with temporal prepositions such as na (E. after) and op (E. on). These prepositions
function as fillers for slots that are highly relevant for this particular meaning of koffie.
As they form combinations of function words plus lexical words they are considered
to be grammatical, not lexical, collocations. A highly relevant slot (from a conceptual
point of view) in combination with a filler that is difficult to predict (Du. op), here
yields a bound, albeit grammatical, collocation.

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.12 (62)

 Willy Martin

slap/sterk drinken malen kopen

(E. weak/strong) (E. drink) (E. grind) (E. buy)

zetten kopje branden inschenken

(E. make) (E. cup) (E. roast) (E. pour out)

DEGREE OF
BINDING
Token-Bound Type-bound Non-Bound

BOUND FREE

Figure 3. Lexical collocations and combinations with koffie (E. coffee) as overlapping cate-
gories

Token- and type-bound lexical collocations

– koffie zetten (E. make coffee)
– slappe/sterke koffie (E. weak/strong coffee)
– koffie drinken (E. drink coffee)
– (een) kopje koffie (E. (a) cup of coffee)
– koffie branden (E. roast coffee)

All these cases are – at the very least – type-bound in a prominent way. The reason
for this is that in the knowledge frame, koffie (E. coffee) as a drink is most prominently
specified by the following slots: PREPARE (coffee is a drink that is prepared in a certain
way), FLAVOUR (one expects coffee to have a certain prototypical flavour), WAY OF
USING/CONSUMING (coffee is consumed in a certain way), CONTAINER (coffee is
drunk out of a certain specific container). From the above series zetten, slappe/sterke,
and kopje are the most token-bound. Zetten because it is idiosyncratic and highly re-
stricted (see Section 3.2.1), slappe/sterke because its distribution is irregular,9 and kopje
because drinks typically have their own containers in the referential world that is re-
flected in language usage. In English, we drink a glass of milk, a cup of tea or coffee, a
bottle of coke (in Dutch we use the diminutive here: flesje – little bottle) etc. This het-
erogeneity makes it difficult to come up with a general, type-bound, prediction. The
more so, as we also drink coffee out of mugs and plastic beakers but still say that we are
drinking a cup of coffee.
As to koffie drinken (E. drink coffee) this is typically a type-bound combination (see
the discussion in Section 3.2.3). However, as I have argued there, it can ‘intrude’ into
the token-bound class because of the fact that koffie (E. coffee) is one of our prototypical
drinks (see the discussion on prototypicality). Thus, koffie drinken (E. drink coffee)
becomes an in-between or transit case and is situated in the overlapping zone between
token- and type-bound collocations in Figure 3.

. Less common alternatives for slap are, for instance, flauw and flets (E. weak, watery).

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.13 (63)

A unified approach to semantic frames and collocational patterns 

The last example, koffie branden (E. roast coffee) differs from the preceding ones
in as far as it concerns another meaning of koffie, viz. the beans or seeds that have to
be prepared before they can become ‘proper’ ingredients for the drink. In this respect
there is another frame at play that triggers both koffie branden (E. roast coffee) and koffie
malen (E. grind coffee), branden (E. roast) and malen (E. grind) being typical fillers for
the slot PROCESS. The more predictable these fillers are – which is certainly the case
for malen (E. grind) and to a lesser degree also for branden (E. roast)10 – the more
we are dealing here with collocators that are more conceptually rather than lexically
bound, and can therefore be considered to be (more) type-bound.
Free lexical collocations
– koffie inschenken (E. pour out coffee).

As inschenken only has an indirect conceptual link/relationship with koffie, viz. via the
‘CONTAINER’ slot which is presupposed when one pours out a drink, it cannot be
regarded as (strongly) type-bound. Therefore its meaning impact is minor and I will
consider it to be free, although its lexical filler is not fully predictable.11
Free combinations
– koffie kopen (E. buy coffee)12

Although koffie is a well known product that is sold and bought much in the Low
Countries, it is not a lexical collocation at all. Indeed, as kopen no longer can be pre-
dicted by a particular slot in the frame of koffie, no filler can be formulated/expected
here. Instead koffie kopen (E. buy coffee) is an example of a free combination between
kopen (E. buy) and koffie (E. coffee), the latter functioning as a non-prototypical filler
in the ‘GOODS’ slot of kopen.

. Concluding remarks and further prospects

In this chapter I have tried to make clear why not only word meanings but also
word combinations, in particular, collocations can profit from one and the same
frame-based approach. In doing so I have stressed both the representational aspects

. In Dutch we use roosteren (E. roast) with pinda’s (E. peanuts) and branden (E. roast) with
amandelen (E. almonds) for comparable processes.
. Instead of inschenken, uitschenken is also possible (cf. E. pour out) but there is a difference in
perspective. Yet, as often is the case with ‘alternatives’, there might also be a semantic difference
as uitschenken often has a resultative aspect: ‘pour out until there is no more left/ until the
container is empty’. This might have an impact on the choice. Anyway, in Dutch we prefer to say
inschenken, at least with koffie.
. This ‘free’ example does not occur in the RBN under the entry koffie.

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.14 (64)

 Willy Martin

of collocations by introducing a new model, and the acquisitional aspects by using

frames both as an analytical instrument and as a heuristic device.
The new model integrates rules, constraints, preferences and prototypes in its de-
scriptive apparatus and thus enhances the purely quantitative, statistical, corpus-based
approach with a more qualitative, cognitive, frame-based one. This approach does
not exclude the quantitative approach but complements it. So, for instance, frames
as analytical instruments allow us to interpret/structure/analyse collocational candi-
dates as found in corpora or gathered by intuition. Frames, in other words, present a
framework for reading the data (see the koffie examples). The fact that they do so in a
systematic way, using ‘degree of binding’ with slots and fillers as a guiding principle,
leads not only to greater consistency but also to greater explanatory power for the data
under study (motivation of the combinatorial data).
As a heuristic device, frames offer the possibility of superseding the data as found
in corpora, by intuition or otherwise. In other words, collocational patterns as gener-
ated by frames act as expectation patterns, not only for drawing the attention to what
(frequently) occurs, but also to what does not occur. In this respect the ‘generation’ of
these patterns is at least as important as the selection of lexical collocations as such, as
the former is a pre-requisite to the latter.
However, it will have become clear that, in order to make the frame-based ap-
proach fully operational, three other things are needed:
1. First, more empirical research is needed in order to build a frame-based lexicon
with types, subtypes, frames, slots and fillers. Failing this, a more heuristics-based
approach will have to be followed for most languages for the time being.
2. Second, in order to test and validate the relevance of the collocational model/hy-
pothesis presented, psycholinguistic research is needed.
3. When it comes to applications of the approach, be they in language didactics or
in language technology, an adapted user model will be needed in order to find out
what is redundant/informative given the knowledge level of the intended users
(native speakers, machines, second language users, etc.).

Acknowledgements

I would like to thank Dr Hennie van der Vliet and Drs Jeroen Redel for our productive
discussions on this subject. I am, of course, solely responsible for all possible errors or
misinterpretations.

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
 TSL[v.20020404] Prn:31/03/2008; 10:20 F: Z13903.tex / p.15 (65)

A unified approach to semantic frames and collocational patterns 

References

Croft, W. & D. A. Cruse (2004). Cognitive Linguistics. Cambridge: Cambridge University Press.
Feilke, H. (2004). Kontext – Zeichen – Kompetenz. Wortverbindungen unter sprach-
theoretischem Aspekt. In Steyer, K. (ed.) Wortverbindungen – mehr oder weniger fest, 41–64.
Berlin: De Gruyter.
Fillmore, C. (1977). Scenes-and-frames semantics. In Zampolli, A. (ed.) Linguistic Structures
Processing, 55–81. Amsterdam: North Holland.
Fillmore, C. (2003). Double-decker definitions: The role of frames in meaning explanations.
Sign Language Studies 3(3): 263–295.
Fillmore, C., C. Johnson & M. Petruck (2003). Background to FrameNet. International Journal
of Lexicography 16(3): 235–250.
Hausmann, F. J. (2004). Was sind eigentlich Kollokationen? In Steyer, K. (ed.) Wortverbin-
dungen – mehr oder weniger fest, 309–334. Berlin: De Gruyter.
Kahneman, D. & A. Tversky (eds.) (2000). Choices, Values and Frames. Cambridge: Cambridge
University Press.
Martin, W. (2003). Definitions and collocations in dictionaries. Lexicographica. Series Maior
113: 3–23.
Martin, W. (2006). Een kwestie van frames. Valedictory Lecture, Vrije Universiteit Amsterdam.
Martin, W. (2007). The lexicon is a (kind of) frame. In Miyares, L.R. & al. (eds.) Actas X simposio
internacional communicación social, 410–418. Santiago de Cuba: Centro de Linguistica
Aplicada.
Martin, W. & I. Maks (2005). Referentie Bestand Nederlands. Documentatie. Amsterdam: Vrije
Universiteit, Research Group Lexicology/Terminology.
Mel’čuk, I. (1998). Collocations and lexical functions. In Cowie, A. (ed.) Phraseology. Theory,
Analysis and Applications, 23–53. Oxford: Clarendon Press.
Mel’čuk, I. & A. Zholkovsky (1988). The explanatory combinatorial dictionary. In Evens, M.
(ed.) Relational Models of the Lexicon, 41–74. Cambridge: Cambridge University Press.
Minsky, M. (1977). Frame-system theory. In Johnson-Laird, P. N. & P. C. Watson (eds.)
Thinking: Readings in Cognitive Science, 355–376. Cambridge: Cambridge University Press.
Poulsen, S. (2005). Collocation as a Language Resource. A Functional and Cognitive Study in
English Phraseology. PhD dissertation, University of Southern Denmark.
Schmitt, N. & N. Carter (2004). Formulaic sequences in action: An introduction. In Schmitt, N.
(ed.) Formulaic Sequences, 1–22. Amsterdam: John Benjamins.
Steyer, K. (ed.) (2004). Wortverbindungen – mehr oder weniger fest. Berlin: De Gruyter.
Wermuth, C. (2005). Een framegebaseerde benadering van classificatiesrubrieken: cardio-
vasculaire rubrieken als case study. PhD dissertation, Vrije Universiteit Amsterdam.

EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use
EBSCOhost - printed on 1/20/2021 6:33 AM via YONSEI UNIV. All use subject to https://www.ebsco.com/terms-of-use