Combination, Collocation and Multi-Word Units 1 Syntagmas

APPROACHES TO LEXICAL COMBINATORICS

Combination, Collocation and Multi-Word Units

František

CERMÁK, Praha, Czech Republic

Abstract

There is a profusion of labels used for the phenomenon of multiword units (MU’s), yet they are notori-

ously underrepresented in handbooks of both grammatical and dictionary type alike. After an illustration

of both, problems of criteria which delimit MU’s are raised up, together with an number of open issues

and necessity of a functional approach is stressed. As three major issues, stableness, typicality and po-

tentiality are discussed. A number of urgent questions waiting for a systematic solutions is listed and a

ﬁnal plea for a more balanced selection and approach is made.

1 Syntagmas, Combinations, Collocations and Other

In the multitude of approaches and handbooks a number of labels, both familiar and less fa-

miliar, is used to designate various syntagmas and word combinations that one comes across.

If anything may be pointed out in these in general, then it is a lack of consistency of many

types to be found here. In the situation of general consensus lacking and conﬂicting theories

and views being proposed, this situation is not surprising. What is surprising is that the same or

very similar phenomena are given widely different names even in the same single approach and

book.

1.1 Grammatical Approaches

Grammatical approaches often tend to play the multiword units (MU’s) down and neglect them.

For a ﬁrst illustration of this, let us have a look, at a representative grammar work, a modern

classic now, namely Comprehensive Grammar of the English Language (Longman 1985) by

R. Quirk, S. Greenbaum, G. Leech and J. Svartvik. It seems that a number of rather different

approaches have been applied here. There are three main areas where multiword units have been

observed or, rather, noted to exist, namely, verbs, prepositions and, up to some extent, nouns.

The respective verbs are illustrated by such casese as drink up, dispose of, get away with and

are called multi-word verbs (pp. 1150ff.) here. These are then further subclassiﬁed as phrasal

verbs, prepositional verbs and phrasal-prepositional verbs, respectively. All of these cases are

said to be a class of words which behave as single units. The apparent contradiction is disposed

of by a somewhat unorthodox argument:

"The term "word" is frequently used, however, not only for a morphologically de-

ﬁned word classes, but also for an item which acts as a single word lexically or

syntactically... It is this extented sense of ’verb’ as a ’unit which behaves to some

extent either lexically or syntactically as a single verb’ that we use in labels such

as ’prepositional verb’".

489

Proceedings of EURALEX 2000

Now, while the argument is true, the label "word" is difﬁcult to accept here for a combination of

words. Is is either one word or two words here, tertium non datur. One may only wonder why

the term lexeme or, rather, multi-word lexeme is avoided.

Only a brief attention is paid to idioms here which are said to differ from multi-word verbs

in types of transformations they might undergo (pp. 1162ff.). It is difﬁcult to accept the ﬁrst

"semantic" criterion mentioned for the idiomatic status of MU’s, namely a frequent existence of

a single-word counterpart, e.g. call for - visit, as too imprecise, non-systematic and broad. On

the other hand, the second criterion, that of the meaning being unpredictable, oddly contrasts

with examples given in its support, such as chatter away, ﬁre away, work away. Without going

into technicalities, it is difﬁcult to see these as really having a fully unpredictable meaning.

Multi-word prepositions, such as apart from, as for, due to, thanks to, is an obvious class of

of MU’s. Yet, it has been given here a different name, namely that of complex prepositions.

While the prepositions are said here to be indivisible both syntactically and semantically, their

generally open character is acknowledged and the boundary between a complex preposition and

a single one is considered to be uncertain and open (pp. 669ff.). No mention is made, however,

of idioms in this respect.

A similar situation is to be found with nouns, too, in that no mention is made here of idioms,

although their frozen character is acknowledged. Moreover, no multiword character of such

nouns is mentioned here either and the phenomenon is handled under the traditional label of

compounds only (pp. 280ff.). The examples given here include such as assistant director, break-

down, sit-in. This terminological inconsistency is further enhanced by a subsequent introduc-

tion, without any additional characterization, of the label "parallel structures" for some other

types of multiword nouns. These are being illustrated, however, by rather typical idiomatic ex-

amples (although the term idiom is not mentioned), such as arm in arm, side by side. Somewhat

later, ﬁnally, and as if as an afterthought, adverbial phrase idioms are mentioned, too, illus-

trated by face to face, but not related to the preceding parallel structures. One cannot escape the

impression of neglect, hesitancy and no ﬁrm policy here, unfortunately.

It is disconcerting to see how capricious grammarians can get, as, apart from an occasional re-

mark of adverbs such as by day, by night (p. 688), there is no mention whatsoever made here

of MU’s in chapters dealing with other word classes. It might be contended that their incon-

sistent choice of phenomena they decided to treat is due to the bulk of phnenomena relegated

to and covered by dictionaries (although the fact is never mentioned). Yet, one must wonder

why grammarians pick up for treatment only s o m e combinations and leave unattended others,

equally important, and having the same character?

1.2 Dictionary Approaches

It is not surprising, then, that also dictionaries, almost invariably, give an equal impression

of being inconsistent in their attention paid to MU’s. There have been two basic approaches,

forming opposite ends of a scale: one listing MU’s under the chosen sense of the single-word

lemma, and another one listing them separately at the end of the dictionary article. The ﬁrst

approach, perhaps an older one, tries to pin down, or rather, guess which sense of the single-

word lemma might correspond to a particular MU, grouping, then, MU’s there. This seems to

490

APPROACHES TO LEXICAL COMBINATORICS

be an attempt almost metaphysical in its very root. As a consequence, it is difﬁcult to ﬁnd MU’s

in any reliable way as this guessing game is rather difﬁcult to follow (and appreciate). This is

patently obvious in particularly large dictionary articles.

The second approach, not pretending to know how to classify MU’s according to senses, op-

erates at the end of the dictionary article. Generally speaking, it is difﬁcult to accept the un-

derlying conviction that each idiom (MU) has somehow a sense directly attributable to a sense

of the single word lemma. Hence, two alternative variant approaches of this can be found, one

simply listing MU’s under a common label at the end of the article and one, also at the end of

the article, listing MU’s in some kind of an order giving each a separate number. The former

variant approach has been adopted by, for example, New Oxford Dictionary of English (1998),

the latter by Random House Websters Unabridged Dictionary (1996).

As a rule, there is no particularly reliable criterion applied in the ordering of MU’s in the ﬁrst

approach, neither in their functional labelling and the user has no way of knowing, except for

their deﬁnition, how the MU’s are used. An odd decision has been applied in the case of some

nominal MU’s which are not handled in this section (headed as PHRASES) but as separate

lemma’s in the alphabet (e.g. blue shark, blue shift, blue chipper). Disregarding the fact that the

order introduced in the second approach is equally a problem, a major inconsistency is in the

false impression evoked by the sequential numbering given both to individual senses of single-

word lemma and to individual MU’s. It is difﬁcult to accept the underlying impression that each

MU represents a separate sense of the headword.

Also here, no attempt is made to distinguish any further, such as to offer a functional labelling

to MU’s. It may be useful for the user to know, next to what the MU means, also how it is

used. One of the ways how to achieve this, is to provide it with a functional information on, for

example, word class type. It would help the user to realize in what ways a MU can and should be

incorporated into a sentence. Elsewhere (

Cermák 1998), it has been shown that idioms seem to

be copying, in their behaviour, the word class function of the single word lexemes and that there

are as many functional types of idioms as the number of word classes. Thus it is not difﬁcult

to adopt a uniﬁed approach to MU’s and provide them with this information, too. Consider, for

example, such cases, as being idioms or, rather, MU’s, as multiword conjunctions, prepositions,

particles, interjections and adverbs (to leave aside the more notorious ones, such as verbs and

nouns), as illustrated by as if, even though (: conjunctions); as to, as long as (: prepositions);

all right, as well (: particles); All right!, For God’s sake! (: interjections); for good, on the other

hand, from head to foot (: adverbs) etc.

2 Problems of Criteria and Open Questions

2.1 Syntagmatic aspects

Providing MU’s in dictionaries with functional labels may not be the most important problem

to be solved. There are other problems as well. Most dictionaries do not even attempt to strike

some sort of balance between coverage of their paradigmatic and syntagmatic aspects and it is

the syntagmatic aspects which are sadly lacking in what might be called a decent representation.

While some dictionaries do not even attempt to state that the noun attention collocates, among

491

Proceedings of EURALEX 2000

other things, with both the verbs pay and give, some hide the fact in sentential examples, while

some other merely list the combinations. Listing them is about most what one can expect to get

from even a large dictionary. Even specialized works, such as The BBI Combinatory Dictionary

of English (BBI 1986), never tell one what the difference in the use and meaning of both might

be and may be somewhat misleading. Thus, The Word Bank corpus gives 79 combinations of

attention with forms of pay, while the number of its combinations with give, i.e. 35, is 50 per

cent smaller. Yet it is given much more prominence in the BBI being listed independently under

a separate number. To take a different example, on the basis of the collocation in broad daylight,

a foreign learner of English might be tempted by the lurking analogy to try to combine broad

with its opposite, too, making it broad night. Yet this is exactly the kind of information which

is never given reliably in dictionaries and the user is left to his or her own devices here. Obvi-

ously, the problem, probably the most serious for a lexicographer, is the degree of stableness of

combinations since stableness is closely linked to acceptability.

2.2 Stableness

It is not very often that the problem of stableness (or stabilization) of lexeme combinations is

brought up, let alone helpfully tackled. People may not even realize that the crucial question to

be ﬁrst asked, to put it linguistically, is "does one deal here, in the given combination, with a

phenomenon of la langue already or still of la parole?". All of the stable combinations, MU’s, are

part of la langue and merely reproduced in speech, whereas combinations which are not stable,

ﬁxed (some prefer to call them preconstructed, frozen etc.) are part of la parole, the speech,

and are formed ad hoc, again and again in each case. So far the theory. The problem with such

a nice clear-cut theory is that it does not always work. Since, obviously, there is a multitude

of combinations which share characteristics of both, one has to use a cline or scalarapproach

here with as many grades between the two opposites as one may ﬁnd and discern. Yet the

necesssity to use this scalar approach does not invalidate the distinction. Rather, it views it

prototypically, in an ideal case, and it is up to the lexicographer to draw the line. However, the

enormous lexicographer’s aid in today’s corpora has its pitfalls, too, in the current overemphasis

on typicality, on inclusion into the dictionary only of what has been sufﬁciently and amply

attested.

One may wonder what kind of beneﬁt could be derived from, for example, a careful selection

and inclusion of current and yet not so typical cases of lexeme combinations stretching the

norm somewhat whereby also less typical but not uncommon collocations would thus indicate

a further and possible use.

2.3 Typicality

In the quest for criteria, if any (with most approaches offering only pragmatical ad hoc solu-

tions), a kind of answer is sought in typicality. While the best solution is seen in the Mutual

Information and t-score so far where the quest for criteria has rested, the problem is far from

being solved. In a recent contribution (Trap-Jensen 1996), a distinction was made between two

kinds of typicality, one provided by the corpus and one by data from a simple association test

in Danish. However, examples such as white minority rule, white coat (coming from the Danish

492

APPROACHES TO LEXICAL COMBINATORICS

corpus) as against white snow, white sheet (of paper) (coming from an association test) seem to

show two different phenomena rather than two kinds of typicality.

While the former examplesstand for typical collocations, it is rather problematic to call the latter

standard collocations (whatever that might mean). Admittedly, these are difﬁcult to come by in

a corpus of any size so far. However, a slightly more common form of the latter of the type, i.e.

snow/paper is white, being used for classiﬁcation of more than one kind, signals rather special

(but hardly typical) occasions when such cases might be used and the particular quality (white

colour) of snow or paper explicitly mentioned. It is obvious that whiteness is a deﬁnitional

quality which might be one of more, perhaps many, and used, for example, for identiﬁcation. It

is very much part of la langue; de Saussure would undoubtedly call it part of valeur, value of the

lexeme. There might be, ultimately, also a cline here without any sharp dividing line between

the two. Yet, it seems, for the time being, that an alternative, "temporal" way of how to view the

two kinds of syntagmas or combinations lies in the stability of the existence of their denotata,

as these are reﬂected by the language system. There is hardly any doubt about the constant,

"timeless" quality of the colour of snow, but there is nothing constant about, say, the colour of

the minority rule. Contrary to what statistics and frequency shows, the degree of stability of the

former in the language is far greater than that of the latter. It is evident that, despite an obvious

interrelationship, such a type of stability is not to be confused with the stableness or, rather,

ﬁxity of a combination in the sense referred to above.

There is, however, room for a possible misunderstanding, too. It would hardly do to consider

either kind to be outside of either la langue or la parole, of either the lexicon as part of la langue,

or the area of semantico-grammatical rule application in texts. There is nothing in between

(contrary to Trap-Jensen 1996, 283) and seeming transitions may turn out to be due to a lack of

data or differences in norm applied by different users.

2.4 Potentiality

Next to positive facts, units and their combinations, there are also rules in the lexicon. It is of-

ten forgotten, while speaking about rules, either semantic or grammatical, that there are other

potential rules, too, in the lexicon, whose existence is directly due to a prior existence of the

former kind and which represent a kind of extension of the former. Disregarding for the mo-

ment that even large corpora are far from ideal, one may become aware of the opposite end

of a new scale here, of obvious potentiality to create, on the basis of a strong analogy, new

combinations which are (almost) perfectly acceptable. While much may be objected against the

recent Chomskyan overblown emphasis on the language creativity, this is not quite the same

thing. Especially in cases where common sense suggests a possibility, speakers do not hesitate

to form a new combination and use it. This potentiality, contrasted to probability, however, has

not found its treatment in dictionaries, so far. After all, not all users wish to speak and appreciate

only what is typical and pre-patterned and they might wish to strike a subjectively new balance

between the new and the stereotype.

2.5 Rules and Regularity

A question often, though rather implicitly, asked by the lexicographer is about regularity of a

given combination. Yet, what is regular, is in fact related to a number of very different rules,

493

Proceedings of EURALEX 2000

some of which holding for a particular and rather narrow domain only, but related also to an

endless debate should semantic rules be included. Thus the seeming commonsense approach

saying that any irregular combination is of (primary) interest to the lexicographer does not say

much, as the distinction regular-irregular is far from being clear.

A different kind of regularity is to be found in usage where standard use is rule-based, of course.

However, every kind of usage means following certain rules to the exclusion of other; there are

always some rules which do not apply. One may, then, wonder, if a kind of guidance to the

usage may not be offered, at least in some instances, by indication, for example, the impossible

or "forbidden" cases. While this interesting problem would ultimately lead to a discussion of

what is feasible, let me point out at least one positive example here instead. On the Czech

dictionary of idioms (

Cermák 1994) this possibility has been succesfully employed by explicitly

indicating a set of (morphological and syntactical) categories or, rather, transformations, which

a given idiom does not normally undergo.

2.6 Outstanding questions

The number of questions waiting for an answer and solution seems to be endless. In connection

to the distinction just made, some of the outstanding question might, brieﬂy, be:

1- How relevant for the identiﬁcation of an acceptable combination is its length, extension

(in number of words, usually), as well as its discontinuous character occurring rather un-

comfortably often? It is natural that most illustrations one ﬁnds in dictionaries are simple

binary adjacent combinations. But one should expect more.

2- How are certain stereotypes and other prebabricated combinations to be identiﬁed and

represented if these have a rather unstable, multivariate form? An answer of a kind is

linked to the size of corpora used. It would seem to be just impossible to ﬁnd (often) two

major variants with an equal frequency. Thus, an invariant is indicated.

3- How are exocentric types of combinations to be identiﬁed and covered? Their character

does not yield the impression of an entity so readily as that of endocentric construc-

tions. Undoubtedly, there is always the lexicographer’s intuition at play while choosing

one example and discarding another one. And intuition seems to be rather in favour of

endocentric combinations.

4- To what extent should combinations, collocations etc. in their treatment be viewed typo-

logically? The fact that, for example, polysynthetic languages rely heavily on compounds

where other languages use combinations should not prevent one from the same effort in

their description. Obviously, discrete character of combinations is due to the typological

character of the language in question. It is difﬁcult to see any difference in criteria and

approach between collocations and compounds, which are, basically, of the same charac-

ter.

494

APPROACHES TO LEXICAL COMBINATORICS

3 Desiderata and Solutions?

It is obvious, as many people would argue, that dictionaries should ideally include a more bal-

anced selection of all system units on a multiword level. However, since there is no such thing

as an ideal dictionary offering this desirable state of affairs one must assume that real, practical

solutions are always based on some kind of non-ideal selection. And it is up the user to either

accept it in good faith (realizing, perhaps, that it is difﬁcult to do better) or (being more suspi-

cious of money-oriented publishers) to distrust it. However, without answers to the above and

other questions it is still very much a wishfull thinking and the practice of ad hoc solutions will

not be abandoned. Feasible solutions have to be sought in the identiﬁcation of criteria which

follow from these questions. Since most of the problems outlined have a scalar character, a spe-

ciﬁc kind of answer here, speciﬁcally that of selection, is to be found, among other things, in

their frequency proﬁles. Resulting varying degrees of selection on a scale could be attributed to

various sizes and types of dictionaries.

References

[1] Benson, M. Benson, E., and Ilson, R. (1986). The BBI Combinatory Dictionary of English. A Guide

to Word Combinations. John Benjamins Publishing Comp., Amsterdam/Philadelphia.

[2]

Cermák, František (1994). Czech Idiom Dictionary. in Euralex 1994-Proceedings, eds. W. Martin,

W. Meijs et al., Euralex Amsterdam, pp. 426-431.

[3]

Cermák, František (1998). Linguistic Units and Text Entities: Theory and Practice, in Actes EU-

RALEX’98 Proceedings, Th. Fontenelle, Ph. Hiligsmann, A. Michiels, A. Moulin, S. Theissen

(eds.). Université de Liege, Liege, pp. 281-290.

[4] The New Oxford Dictionary of English (1998). Oxford University Press, Oxford.

[5] Random House Websters Unabridged Dictionary (1996). Random House, New York.

[6] Trap-Jensen, Lars (1996). Word Relations: Two Kinds of Typicality and Their Place in the Dictio-

nary. in Euralex ’96 Proceedings I-II. University of Göteborg, G"oteborg, pp. 283-291.

495

Proceedings of EURALEX 2000