Vous êtes sur la page 1sur 13

Cognitive Grammar: some basic concepts 121

infer the something else for which it stands. Language is, of course, a

6
sign system par excellence.
There are several ways in which a sign can mean (Taylor 2002: Chapter 3). In the
case of INDEXICAL signs, the perceptible event can be traced back to its presumed cause.
Smoke is a sign of fire: if we observe smoke rising into the air, we presume there is a
Constructions fire which causes the smoke. Medical diagnosis proceeds on the basis of observed
(and reported) symptoms. Some aspects of language are indexical. Slurred speech
might be taken to signify the inebriated state of the speaker; a deep baritone voice
communicates that the speaker is a male of the species; hearing a Scottish accent we
In previous chapters I have referred on several occasions to constructions. The reasonably infer that the speaker is Scottish.
term is common in linguistic discourse. We can refer to the passive constructions, Another way in which signs can mean is though an ICONIC relation between the form of
the relative clause constructions, and the conditional constructions of a language. the sign and its meaning. The sign in some way resembles the meaning which it stands
Students of Latin will recall having learned about the ablative absolute construction. for. Many road traffic signs belong to this category, as do the onomatopoeic words in a
Although familiar, the term is far from uncontroversial. Generative linguists have language. Aspects of text structure may also display iconicity. To mention event A before
tended to dismiss constructions as epiphenomena, the product of more fundamental mentioning event B can convey, iconically, that event A happened before event B.
principles, of no particular interest in themselves. Others, however, take a different Although language may display elements of indexicality and iconicity, language is
view, putting constructions in centre stage and regarding them as one of the most primarily neither indexical nor iconic. Linguistic signs are meaningful in another
basic of all linguistic constructs. In fact, in many construction-based grammars, respect, namely, in virtue of a SYMBOLIC relation between form and meaning. Symbolic
constructions perform the work which in generative theories is performed by rules. signs do not resemble what they mean, neither is there a causal relation between a
There are, as we shall see, a number of different ways in which constructions can sign and what it signifies. The relation, instead, is purely a matter of convention,
be characterized. Thus, something which might be regarded as a construction on one which all users of the sign have to learn and comply with. An important feature
approach might turn out not to be a construction on other approaches. In order to of symbolic signs is that they are under the control of the agent who uses them.
provide some background to a discussion of these different approaches, I first present A person chooses to use the sign, and does so with the intention of representing the
the basic assumptions of Langacker's theory of Cognitive Grammar. Later in the associated meaning.
chapter I address some of the issues arising from an approach to language which Cognitive Grammar is based on the view that language is inherently and quintes-
sentially symbolic. Given this assumption, the study of language involves the study of
gives preeminence to constructions over rules, even to the extent that rules (or at
three kinds of object:
least, rules as traditionally understood) are no longer needed.
(a) the linguistic signs themselves, that is, language in its perceptible form,
Cognitive Grammar: some basic concepts prototypically as sound, but also encompassing written forms and (in the
case of sign languages) hand movements. Following Langacker, I will use the
Langacker began working on the theory of Cognitive Grammar in the mid-197os, in
term PHONOLOGICAL REPRESENTATION to refer to this aspect of language.2
part as a reaction to the generative theories of the time.' The standard presentation is
(b) the symbolized meaning. Meaning is being used here in a rather broad
the two-volume Foundations of Cognitive Grammar (Langacker 1987, 1991); a more
sense, to refer not only to propositional content and referential intent, but also to
recent account is Langacker (2008). In this section I introduce some basic principles the expression of the speaker's attitudes and beliefs. I refer to this aspect
of the theory, especially as these pertain to constructions.
of
Let us begin by considering language in the broader context of communication.
language as the SEMANTIC REPRESENTATION.
Humans (and other animals) communicate by means of signs. A SIGN iS a perceptible
(c) SYMBOLIC RELATIONS between elements of (a) and elements of (b).

For the historical background to Cognitive Grammar, see Taylor (2007). object or 2
Evans (2009) introduced the term vancu to refer to the physical manifestation of a linguistic sign, in
whatever medium.
event which stands for something else. Observing only the sign we can
122 Constructions Cognitive Grammar: some basic concepts 123

Symbolic relations are most clearly illustrated with reference to words. A word such Importantly, the inventory of units which make up a language does not
as tree associates a phonological representation [trid with the concept 'tree'. The constitute a random, unorganized list. Units may be related to each other in various
notion of symbolic relation applies to many other linguistic phenomena, however. ways. Three kinds of relation are important:
Indeed, a distinctive claim of Cognitive Grammar is that all of language can be
(a) the PART—WHOLE RELATION. One unit may be part of a larger unit. Conversely, a
described in terms of the three kinds of entities listed above. Cognitive Grammar is
complex unit can be analysed into its constituent parts. The specification of a unit
therefore a MINIMALIST theory of language, indeed, a RADICALLY MINIMALIST theory.3
may include information regarding the larger units of which it may be a part.
In its insistence on the symbolic nature of language, Cognitive Grammar is at
(b) the SCHEMA—INSTANCE RELATION. This has to do with the abstractness of a unit.
variance with most mainstream theories. Other theories of language certainly
One unit can be SCHEMATIC for units which are specified in more detail.
acknowledge that language is a means for relating sound and meaning. They insist,
Conversely, a unit may INSTANTIATE a more schematically characterized unit.
however, that the relation is not a direct one but is, rather, mediated by intervening
(c) the BASED-ON RELATION. One unit may be structured on the basis of its similarity
levels of organization. These intervening levels have their own distinctive structural
with another unit,
principles, involving elements and relations which are unique to these levels. The
most important of these intervening levels is syntax, which is structured from These relations pertain to phonological, semantic, and symbolic units alike. More-
uniquely syntactic elements such as the lexical categories (noun, verb, etc.), the over, the relations may be recursive. For example, [A] may be part of [B],
phrasal categories (noun phrase, verb phrase), and the various relations which can which in turn is part of [C], or [X} may be an instance of [Y], which in turn is an
hold amongst these elements (such as the relation of subject-of a clause). The instance of [Z].
Cognitive Grammar claim, on the other hand, is that syntax can be fully To get a flavour of how this model might work, consider an example from
described in terms of symbolic relations between phonological structures and phonology. The word-form (six)] is an established phonological unit, associated
semantic structures. with the concept 'sing'. We can analyse the unit into its parts, namely, the segments
On the Cognitive Grammar view, then, a language does not comprise a set of rules [s], [I], and (r)), Each of these constituent units instantiates more schematic units: [s]
for generating well-formed syntactic objects. Rather, it consists of an INVENTORY OF and [13} are instances of [CONSONANT], [I] is an instance of [vown], while the complex
LINGUISTIC UNITS. A UNIT is any element of a language (whether phonological, semantic, form fsii)1 is an instance of the syllable schema [CVC]. The syllable schema ECVC1
or symbolic) which has become established, or ENTRENCHED, in the speaker's mind as a may be divided into its parts, namely, an onset unit, consisting of the initial [C]
result of the frequency of its previous use (whether in production or reception). Since component, and a rhyme unit, consisting of the [VC] component. Accordingly, [sx1)1
units are established as a consequence of usage, the inventory of units which may be analysed in terms of its onset [s} and its rhyme [n31. Units of intermediate
constitute a language is open-ended. Using a language is a matter of drawing on schematicity might also be recognized. Thus, [s] is not only a vowel but also a
the available inventory in order to realize one's present needs. Learning a language short (or lax) vowel as well as a mid-high front vowel; frA is not only a consonant but
is a matter of expanding one's repertoire of linguistic units and becoming expert also instantiates [NASAL CONSONANT] and [VELAR CONSONANT]. A peculiar feature of [Id
in their use. Language learning, on this view, is a life-long process. Even for in English is its restricted distribution. The sound can only occur in syllable-coda
competent adult speakers, the inventory of linguistic units is subject to change. position and only after a short (or lax) vowel; [ilis] and [sin)] are not possible
New units are added as a consequence of entrenchment, while previously syllables in English. These facts can be accommodated by proposing, as part of the
established units may atrophy through lack of use. specification of [01, that the sound can only occur in a syllable rhyme containing a lax
vowel. The impossibility (or, if one will, the ungrammaticality) of [ills] and Esiird
follows from the fact that these forms are not sanctioned by any existing phonological
Cf. Culicover (zoo4: 132): "the most radical form of minimalism in linguistic theory is one that starts schema of English.
from the premise that if you can't 'see' it, that is, if it isn't concrete, it isn't there. What is concrete in the With this background information in mind, let us now turn to the notion of
case of language is meaning and sound." One might, to be sure, take issue with the claim that meanings are
`concrete' and can be 'seen'. Be that as it may, Culicover's notion of 'radical minimalism' would seem to construction.
exclude all structural elements which are not directly relatable to the surface form of an expression. In this
connection, Langacker (1987: 53-4) proposed the CONTENT REQUIREMET, according to which the only
structures permitted in the grammar are those that actually occur in linguistic expressions, or else emerge
by abstraction or categorization of the primary data. Structures which diverge from surface forms in
terms of their constituency or their linear sequence are inadmissible. 4
But not, however, after all of the short vowels. There are in English no syllables terminating in MA.
124 Constructions Constructions 125

the other. For example, the Nom constituent of the [Modifier + Nom] construction
Constructions
may be instantiated by an expression with the structure [Nom and Nom].
The term 'construction' has been used in the linguistics literature in a number of Equally, a Nom constituent of the [Nom and Nom] construction may be instantiated
ways (Langacker 2005; Taylor 2004b). There is, as we shall see, some overlap amongst by an expression with the [Modifier + Nom] structure,
the different uses, in the sense that there are phenomena which would count as
constructions on any of the approaches. Equally, some things which might count as a (b) Constructions as pairings of form and meaning
construction on one of the approaches would be excluded on other understandings of On an alternative approach, constructions are defined, not in terms of the part-whole
the term. relation, but in terms of the symbolic relation between form and meaning. On the
most liberal interpretation of this view, any association of a form and a meaning
(a) Constructions as internally complex entities constitutes a construction. The word tree is a construction, since it associates a
In keeping with the internal make-up of the word (construction is a noun derived concept with a phonological form. The sentence you are now reading is, once
from the verb (to) construct), a construction is any linguistic form which can be again, a construction, in that it associates a phonological (or graphological) form
analysed into its parts. The phonological form [sin] is a phonological construction, in with a meaning. A narrower approach restricts the term to form-meaning pairings
that it can be analysed as a sequence of three sound segments as well as into an onset which have unit status, 'Construction' thereby becomes synonymous with the sym-
constituent and a rhyme constituent. The word singer is a symbolic construction, in bolic unit of Cognitive Grammar. On this more restricted understanding, the sen-
that it can be analysed into its parts, [sing] and [-er] , each with its own meaning and tence you are now reading is not a construction, since this specific combination of
pronunciation. The sentence you are now reading is a construction, in that it can be words and phrases has not become established through usage. Note that this
broken down into its component words and phrases. On a more restrictive view, and approach excludes the possibility of phonological constructions, such as [sin] or
one more in keeping with the traditional use of the term, construction refers, not so the syllable schema [CVC]. A syllable may, of course, feature as the phonological
much to actual words and expressions, but to the patterns by which these are put pole of a symbolic unit. However, considered simply as an element of the sound
together. Constructions, in other words, are schemas for the combination of smaller structure of a language, a syllable is not in itself a symbolic unit.
units. The phonological form [sip] instantiates the syllable construction [CVC]. The A more restrictive view still is to regard as constructions only those form-meaning
word singer instantiates the agentive noun construction [V-er]. We may likewise pairings whose properties (whether formal or semantic) cannot be derived from the
speak of noun phrase constructions, transitive verb constructions, relative clause properties of any other constructions (Goldberg 3.995).5 Prime examples of construc-
tions, on this more restricted view, are the constructional idioms—more precisely,
constructions, and so on. This understanding of constructions was quite widespread
the schematic representations of these idioms—that were discussed in Chapter 4. The
during the middle decades of the twentieth century, especially in the writings of
more the merrier would not be regarded as a construction; constructional status
linguists influenced by Bloomfield. Here is Rulon Wells explaining the different would, however, attach to the schema which the expression instantiates, namely [ma
readings of old men and women in terms of the different constructions which the X-ER THE Y-ER]. Note that on Goldberg's approach, the word sing would be a
word string can instantiate: construction since we cannot predict, from anything else we might know about the
language, that this concept will be symbolized by this particular combination of
In the meaning 'women and old men' the sequence belongs to that construction (noun or
sounds. On the other hand, the word singer might not be regarded as a construction.
noun-phrase + and + noun or noun-phrase) which has the meaning of conjunction; the first
The word certainly has unit status. On the other hand, it could be argued that
noun-phrase belongs to the construction modifier + noun or noun-phrase. But in the meaning
it inherits its phonological and semantic properties from the schematic [V-er]
`old men and old women' the sequence belongs to the construction modifier + noun or noun
construction and from the unit which instantiates the [V] constituent.
phrase; the noun phrase in turn belongs to the construction noun or noun-phrase + and +
noun or noun-phrase. (Wells 1957 [1947]: 195)
Let us use Nom as a cover term for Wells's 'noun or noun phrase'. Wells identifies
two distinct Nom constructions. The one has the structure [Modifier + Nom], the
other may be stated as [Nom and Nom]. These specifications state how a Nom
5 According to Goldberg's 4995: 4) definition, "C is a CONSTRUCTION iff, C is a form-meaning pair
expression can be assembled from its parts (alternatively, analysed into its parts). <Fi, Std such that some aspect of Fi or some aspect of Si is not strictly predictable from C's component
A feature of Wells's account is that the one construction can be embedded into parts or from other previously established constructions".
126 Constructions Constructions or rules? 127

(c) Constructions as units form their plural, namely, by suffixation of {s}. These phenomena, one might
The two approaches outlined above differ in their coverage. As we have seen, suppose, are best handled by rules of the traditional kind. One option, then, might
phonological constructions are impossible on approach (b). On the other hand, be to propose a division of labour between generative rules and a battery of idiomatic
morphologically simplex words, such as sing, are excluded on approach (a). Other- constructions, as argued, for example, by Pinker (1999) on the example of verb
wise, the data covered by the two approaches overlap considerably. It may be no morphology. Thus, in Pinker's account, the rules generate the regular past tense
accident that many authors tend to slip between one and the other understanding forms, by suffixation of ft}, whereas irregulars are learned as such.
of the term.
In the next part of this chapter I compare a rule-based and a construction-based
Taking elements from each of the above approaches, we can propose a third
account of the regularities in a language. Although seemingly equivalent, there are
definition of 'construction', referring to whatever a speaker of a language has specifi- some substantive differences between the two approaches and the predictions which
cally learned. On this definition, singer might well have the status of a construction. they make. The discussion will also allow us to reconsider the nature of rules and how
Although the phonological and semantic properties of the word can be derived they are applied. I then turn to one of the entailments of a construction-based
from its component parts and from the schema whereby they are combined, it is grammar. As noted earlier in this chapter, Langacker's radically minimalist
unlikely that speakers who use the word do in fact put it together from its parts. The theory of language brings with it the rejection of syntax as a distinct level of
word is stored, and accessed, as a whole. Similarly, How old are you? is likely to have organization mediating between phonological and semantic representations, (Recall
the status of a construction since this phrase is available as a ready-made whole; it the claim that the only objects of study in Cognitive Grammar are phonological
does not have to be assembled in accordance with a more schematic interrogative representations, semantic representations, and symbolic relations between phonological
construction. `Construction' thus becomes synonymous with the Cognitive and semantic representations.) The very fact of syntactic structure could therefore be
Grammar 'unit'. In brief, a construction is any element of a language that has a crucial argument against one of the underlying assumptions of Langacker's theory.
been learned and that forms part of a speaker's linguistic knowledge. 1 shall argue, on the contrary, that far from threatening the theory, syntax in fact
As noted earlier in this chapter, it is axiomatic in Cognitive Grammar that a provides a very strong argument for the primacy of constructions.
language can be exhaustively described in terms of established units (or, if one will,
constructions), whether semantic, phonological, or symbolic. There is, in this
scheme of things, no place for rules—or at least, no place for rules as they are Constructions or rules?
commonly understood, namely, as instructions for the composition of larger With respect to the regular expressions in a language, the sceptical reader may be
configurations. wondering whether a construction-based account is little more than a notational
There are undoubtedly many phenomena in a language which are best analysed in variant of a rule-based approach. On the constructionist account, we might represent
terms of constructions. These include idiomatic expressions of various kinds, in one type of noun phrase by means of the schematic unit [Det N]. Exactly this same
particular the constructional idioms discussed in Chapter 4 and elsewhere. Construc- structure could be generated by means of a re-write rule of the kind NP -> Det N. On
tional idioms, it will be recalled, are idiomatic in the sense that they cannot be the constructionist account, the expression a rabbit is sanctioned by the fact that a
generated by general syntactic rules. Moreover, the items which are able to fill their instantiates [Det] while rabbit is an instance of [N]. On the rule-based account, a and
various slots can often not be brought under any of the generally recognized syntactic rabbit can slot into the rule output in virtue of the fact that a is marked in the lexicon
categories, and a construction's meaning, more often than not, cannot be derived as a determiner, rabbit as a count noun. The reader may also have noticed how the
compositionally. At the same time, the constructions are productive to a greater or rather tortuous formulations that Wells employed in the above citation (see p. 124)
can be easily, and perhaps more perspicaciously, expressed in phrase-structure terms.
lesser extent, in that new instances can be created in accordance with the construc-
Essentially, Wells was proposing that old men and women can be parsed in two
tional schema. It is worth noting that a major impetus for the development of
alternative ways. Again using Nom to cover Wells's 'noun or noun phrase', we can
construction-based grammars came from the study of constructional idioms in the represent the two structures as follows:
198os and 19908 (Fillmore, Kay, and O'Connor 1988; Kay and Fillmore 1999).
For other phenomena, however, the need for a construction-based account is less
evident. In spite of the ubiquity of the idiomatic and the idiosyncratic—a feature of (1) a. [Nom[Namold men] and [N„„womeni] b.
language that I have emphasized in the preceding chapters—there is still, one might [Nom[Adiold [Nommen and women] ]1
suppose, a place for generative rules. The farmer shot a rabbit is not obviously
idiomatic, nor is the process whereby the overwhelming majority of count nouns
118 Constructions Constructions or rules? 129

Each of these structures could be generated by a set of rules for the formation of Nom Take the case of nouns and their plurals. In principle, every count noun can be
phrases. The rules would include the following: pluralized. The pluralization rule, in English, is very simple: add {s} to the singular
form, The matter is complicated somewhat by the fact that there are a couple of
(z) a. Nom -> Adj Nom dozen exceptions to this rule. Exceptional plurals such as men, children, and mice
b. Nom -> Nom and Nom must obviously be listed and must be learned as such by all speakers of a language.
The question arises, though, whether we also need to count, as part of a person's
The ambiguity of old men and women follows from the ordering of the rules and the
linguistic knowledge, those plurals which can be formed according to the rule. Do we
way in which the one 'feeds' into the other. If (b) is applied after (a), we have the want to say that a person who knows how to form the plural of the noun eye also
structure jAdj [Nom and Nom] ]; applying (a) after (b) we can generate the stores in memory the plural form eyes? If our aim is to minimize redundancy in
structure [ [Adj Nom] and [Nom] 1. (Many other outcomes can of course be the grammar, the answer would have to be 'no'. There are, however, good reasons for
envisaged, due to the fact that the rules can be applied recursively,) believing that in the case of high-frequency plurals at least, these are indeed stored
Again in the spirit of re-write rules, we could represent the content of the syllable as such.
construction [CVO as follows:
Consider the matter from the point of view of acquisition. Although the pluraliza-
(3) a. Syllable -> Onset Rhyme tion rule is very simple, it still has to be learned and this can only happen on the
b . On s et -> C basis of an array of examples. It is only after having noticed that plural nouns (in the
c . Rhyme -> VC
main) end in {s} that the learner is able to hit upon the rule. For the learning process
Additional rules might specify that IC] could be re-written as [pi, [s], and so on, to be possible at all, the learner has to have an array of plural forms available in
while [V] could be rewritten as [ii], [I], etc. memory. On the no-redundancy approach, we should have to assume that the
Given the seeming equivalence of rules and constructions, it is legitimate to ask moment the rule is abstracted the instances on which it is based are immediately
whether there is any substantive difference between the two approaches. Is a con- erased from memory. This is somewhat implausible. More likely, the already learned
structionist account simply a notational variant of the generative account? If con- plural forms, especially if they are in frequent use, are retained.
structions are nothing other than the output of generative rules, there may be no Support for this view comes from LEXICAL DECISION TASKS. On a lexical decision
compelling reason to prefer a constructionist account, at least with respect to fully task, a subject is presented with a sequence of letters and must decide, as quickly
regular, non-idiomatic expressions. Constructions, as Chomsky (1995: 170 has as possibly, whether the sequence constitutes a word or not. Several factors
stated, would be little more than epiphenomena, the secondary consequence of influence the speed of the decision, one being the frequency of the word in the
rules operating over a lexicon. language. Consider, now, recognition speeds for singulars and plurals. If plurals are
Although the two approaches might be descriptively equivalent, in the sense that always generated from the singular, the recognition speed for plurals should
they are able to accommodate the same range of data, they differ with respect to correlate exactly with the recognition speed for singulars. Sereno and Jongman
what a speaker of a language is claimed to have learned. A construction-based (1997) report that this is not the case (see also Alegre and Gordon 1999). Some
approach tolerates a good deal of REDUNDANCY. The same fact about a language may be nouns, such as kitchen and village, tend to be used predominantly in the singular.
stated in more than one place. A pattern may be stated in a schematic construction; Other nouns, such as window and expense, are biased towards the plural. Sereno
the same pattern is manifest in each of the established instances of the schema. The and Jongman found that the recognition speed of plurals reflected the frequency
generative model, on the other hand, strives to eliminate redundancy from the of the plural forms, rather than the frequency of the singulars. The finding entails
grammar. If a form can be generated by rule it does not need to be listed. The only that high-frequency plurals are stored and accessed as such; they do not need to be
things that need to be listed are the irregular and non-predictable forms of a language generated by the pluralization rule, nor analysed into their component parts, on
(see pp. 34-5). However, to the extent that knowledge of a language does contain each occasion of their use.'
redundancies, the construction-based account might be psychologically more
plausible.6 7
It is also worth noting that the singular and plural forms of a noun might have their own characteristic
distributions in the language. As Sinclair (2004: 31) observed, plural eyes typically has a literal, referential
sense (as when we say that a person `has brown eyes'), whereas singular eye commonly has a metonymic or
Langacker (1987: 29) in this context speaks of the RULE/LIST FALLACY. This refers to the (as he claims,
6

metaphorical sense, often in association with fixed expressions (keep an eye on, have an eye for, in the
unwarranted) assumption that if a specific statement can be derived from a more general statement, the
specific statement does not need to be listed.
130 Constructions Constructions or rules? 131

The decision paradigm has also shown that speakers are sensitive to the overall Sosa and MacFarlane (2002) showed that the degree of entrenchment of a larger
frequency, not only of words, but also of word combinations. In a recent study, expression, as measured by its frequency of occurrence, delayed the recognition of a
Arnon and Snider (2010) displayed four-word strings on a screen and asked component word. Subjects were asked to listen out for the word of When the word
subjects to decide as quickly as possible whether the strings were grammatical. More occurred in high-frequency combinations, such as kind of, some of and out of
frequent combinations (as determined by corpus data), such as don't have to worry, responses on a monitoring task were longer than with less common sequences,
elicited faster responses than less frequent strings, such as don't have to wait.8 The such as sense of, example of and care of. It seems that the high-frequency phrases
effect cannot be accounted for in terms of the higher frequency of worry vis-a-vis had been stored as such, rendering their internal analysis less salient.
wait (in fact, according to the BNC, wait occurs more frequently in the language
Other experimental results point to a similar conclusion. Wheeler and Schumsky
than worry), nor indeed in terms of the higher frequency of to worry vis-a-vis to
(198o) asked subjects to divide written words up into their constituent parts. They
wait (again, BNC data indicate that to wait is the more frequent combination). Nor is
report that about half their subjects failed to mark any internal divisions in the word
it a matter of the plausibility of the semantic content of the two expressions.
According to ratings elicited by the researchers, both strings were judged to be equally baker, suggesting that for these subjects, at least, the word had become so
plausible. Rather, it would seem, subjects are responding according to the entrenched as a lexical unit that it was no longer felt to be related to the base verb
frequency of the four-word phrases as a whole, as laid down in memory on the basis of (to) bake, neither was the final -er suffix recognized as such. Consider, also, work
their previous experience with the language. reported in Nordquist (2004). She presented subjects with isolated words and asked
them to construct, for each word, three sentences containing the word. One of the
These findings suggest that frequently used items, whether internally complex words tested was the pronoun I. Corpus data indicate that the personal pronoun is
words or multi-word units, tend to be accessed as wholes, even though, in principle, typically used in expressions such as I think, I guess, and I suppose. These high-
they can be analysed into their component parts. The other side of the coin is that the frequency phrases tended not to be elicited in the experiment, precisely because,
component parts of a complex unit may not be immediately apparent to language according to Nordquist, they are stored and accessed as wholes, rather than being
users. The matter has been investigated using the MONITORING paradigm, where compiled compositionally. One of the sentences proffered by her subjects was the rather
subjects are asked to listen out for a particular item and to press a key the moment bizarre combination I cough, a phrase which had evidently been put together on the
they hear it (Connine and Titone 1996). Several studies have shown that sound basis of the schematic [NP V] schema.
segments take more time to identify in the flow of speech than the syllables of Evidence that regularly formed expressions may be stored in memory also comes
which they are parts, and that syllables take more time to identify than the words from within the language itself. Internally complex expressions, once they have
of which they are parts (Foss and Swinney 1973; Savin and Bever 1970; Segui, acquired unit status, are liable to undergo what Bybee (1985) has referred to as
Frauenfelder, and Mehler 1981). In some respects, this finding is paradoxical. One DRIFT (see also Taylor 2002: 310-11). Although such expressions can in principle be

might suppose that in order to identify the word cactus, a listener must identify the generated by rule, they tend to be stored and accessed as wholes. As a consequence,
first syllable as [kaak], which in turn requires that the first sound of the syllable is they are liable to take on a life of their own, independent of their compositional
identified as [k]. One might expect that monitoring times would reflect this compo- value. I mentioned above that singer is assembled in accordance with the agentive
sitional sequence. The fact that it does not suggests that it is the larger, entrenched noun schema [V-er]. This account may not be strictly accurate. Singer does not
units that are more immediately available for processing than the parts of which they simply refer to tone who sings' (the compositional value). It comes to be used only of
are composed. persons who sing well, competently, or professionally. The word has acquired a
meaning which is not fully predictable from its compositional structure. Likewise,
baker does not simply denote 'one who bakes (things)' (its compositional
mind's eye, in the public eye, more than meets the eye, etc.). Also from a language-internal perspective, meaning). The word can be used of a retailer who sells baked products, such as
therefore, there are good reasons for supposing that the singular and the plural forms of the noun are bread and cakes (not, though, potatoes), irrespective of whether he has himself
associated with distinct representations.
baked these products. Indeed, as noted above, some speakers apparently do not
8
Comparable endings are reported by Bod (2000; cited in Bad 2006: 316) and by Bonnard and
Matthews (2008). Bod showed that frequently occurring sentences, such as I like it, are recognized as even perceive the word to be an agentive noun at all. Murderer, on the other hand,
grammatical more quickly than less frequent sentences, such as I keep it, even after plausibility and word refers to any person who has committed the crime: you do not have to be a competent
frequency are taken into account. Bannard and Matthews found that two- and three-year-olds were able to
or professional killer to be named by this word. Often-used phrases can also acquire
repeat multi-word sequences that had frequently occurred in child-directed speech, such as When we go
out, more quickly and more accurately than tow-frequency combinations, such as When we go in. specialized meanings, over
132 Constructions Applying a rule: What kind of process is it? 133

and above, or even at variance with their compositional value. Would you believe it? is construction, in which the experiencer (the thinker) is expressed by dative me while
not really a yes-no question, enquiring whether the addressee would, or would not, the verb takes the third person singular inflection.
believe something. It is more an expression of speaker surprise and does not require a The requirement for speakers to generate each and every complex expression that
`yes' or `no' response. I don't know can be used not only to express the speaker's
they utter, applying the rules of grammar to items from the lexicon, is implausible
ignorance of some matter, but as a kind of hedge, conveying lack of commitment.
from another perspective: it would severely compromise the fluency of speech.
Drift also characterizes phonological structure, in that frequently used combina- Speakers have at their disposal a vast repertoire of ready-made and semi-precon-
tions tend to exhibit a greater degree of phonological reduction, involving assimila- structed expressions, which they can reel off without the need to assemble them, even
tions, elisions, and the like, than less frequent combinations or combinations
though the expressions may be perfectly regular in terms of their internal make-up.
which are assembled ad hoc. I don't know, especially when used in its hedging
You wish to gently discourage someone from some course of action. There is a ready-
function, typically shows reduction of don't to [r9], or even to (D) (Bybee and Scheibman
made expression for this: I wouldn't do that if I were you. You find fault with the way
1999). Palatalization of Edj) to (c131 is more likely to occur in the frequently occurring
phrase did you [did jut] -> [didat4 than in the incidental combination had yesterday your interlocutor has expressed himself, but you don't want to be too direct. Again, a
(Bush zoos). Krug (2003) observed a similar phenomenon in the written language. fixed expression is available: I wouldn't put it like that. You want to express your
High-frequency strings such as I have and you have are often written as I've and indifference; out comes another ready-made formula: I couldn't care less. In sum-
you've; elision is less common with less frequent strings such as they have, there mary, there is evidence from various sources that speakers do learn frequently
have, and where have. In fact, Krug discovered that the frequency of a string encountered forms and have recourse to these in their day-to-day use of the language.
correlated quite well with the frequency of the elided spelling.
Applying a rule: What kind of process is it?
The converse of drift—what we might call PERSISTENCE—can also be observed in
many places in the grammar? Frequently used forms may fail to undergo changes As many researchers have emphasized (Pawley 1985; Pawley and Syder 1983; Sinclair
which are operative elsewhere in the language: their very frequency (and their 1991, amongst others), quite a lot of what people say is formulaic in nature. It is
corresponding entrenchment) protects them from change. In earlier stages of equally clear that not everything that peoPle say consists in the parroting of ready-
English, verbs could be negated by means of not (they sang M they sang not) and made formulae. Speakers are creative and do come up with forms which have not
interrogative forms required subject-verb inversion (they sang sang they?). Nowa- been learned as such. Here, one might suppose, the rule-based approach will come
days, so-called do-support is virtually obligatory for negated and interrogative forms: into its own. Should you have occasion to talk about more than one portcullis, you
They sang they did not sing N did they sing? (Modal verbs are, of course, the will need to come up with the plural form of the noun. Given that the noun is rather
exception: He can sing he cannot sing N can he sing? -a fact which is probably not infrequent, and its plural more infrequent still, it is highly improbable that you are
unrelated to their high frequency: Bybee and Thompson 1997.1°) Even now, however, able to retrieve the ready-made plural from memory. You have to create it.
one occasionally encounters expressions which have held out against do-support.
Applying a rule is often thought of as an operation performed on an input. With
The greeting formula How goes it? is recorded twice in the BNC. I think not is often
respect to pluralization, the input is the singular form and the operation consists
used instead of the expected I don't think (so). Go and think are amongst the most
in adding (s). There is, however, another way to conceptualize the process. The task
frequent of the English wrbs.n An especially curious relic is the expression methinks,
is to come up with an output which conforms with the plural construction, namely
(N. {s)].
On the face of it, the two mechanisms appear to be exactly equivalent. Whether
9
'Persistence' is being used in a slightly different sense from Hopper (i991). you add {s} to an input form, or aim to produce an output which consists of the base

The usage of some of the less frequent modals, such as dare and need, oscillates between forms with noun plus {s}, amounts to much the same thing. In order to be able to differentiate
and without do. Alongside you needn't go we have You don't need to go. Past tense forms of dare tend to
the two approaches we need to turn to cases where a speaker innovates with respect
require do-support: I didn't dare (to) go, while present tense forms do not: I daren't do it. Note also the
fossilized form How dare you!, uttered as a challenge to the addressee. *How do you dare? would be totally to the prevailing norms of the language, or, if one will, where a speaker 'makes a
ruled out. mistake'.
Frequency, however, cannot account for the jocular I kid you not, with eleven attestations in the BNC,
Possibly, the archaic syntax is being used in order to enhance the jocular character of the expression. One type of innovation consists in the regularization of an irregular form. Several
Significant, though, is the fact that the expression is virtually frozen: no other subject or direct object NP is English nouns have a mildly irregular plural form, in which the final voiceless
allowed, neither can the expression be used in the past tense. There are no attestations of They kidded)
him not, and the like. in the sense 'I think'. This harks back to a now archaic impersonal
134 Constructions . Applying a rule: What kind of process is it? 135

fricative of the singular is replaced by its voiced counterpart. Thus, the plural of For a further illustration of the role of output schemas, consider the irregular past
thief is thieves instead of the expected thief's. House is another such noun (though in tense forms of certain verbs (Bybee and Slobin 1982; Bybee and Moder 1983). There
this case the irregularity is not represented in the spelling). Children often fail to is a small set of verbs in English which are unchanged in the past tense (put, cut, set,
produce the standard plural [hauzaz], instead forming the plural in the 'regular' let, and, for some speakers, in some of its uses, fit). These are verbs which already
way: [hausaa Their mistake can be explained equally well on either of the two terminate in a typical exponent of the past tense, namely [t]. It is as if the verbs
approaches. On the rule-based account, we might suppose that the speaker's already 'look' to be past tense forms; hence, no further inflection is necessary. For a
lexicon has failed to tag the noun house as an exception to the plural rule; more complex example, consider the following mildly irregular past tense forms.
alternatively, that the speaker has inadvertently overlooked the tag. On a (4) a. dwell dwelt spell r-‘, spelt
constructionist approach, we should say that the irregular plural has not yet become b. bend bent spend N spent
entrenched; the speaker therefore must fall back on the plural schema. c . keep — kept creep N crept
More interesting, from our perspective, is the reverse phenomenon, whereby a d . feel felt dream dreamt
regular form is replaced by an irregular. One such case of irregularization was
e. leave left cleave N cleft
mentioned in Chapter 1.12 Alongside ['prausesaz] —the 'regular' plural of process—
one sometimes hears, especially in academic discourse, the form Pprausesilzi, with a We may discern a number of distinct patterns here.
tense [ii] vowel in the final syllable rather than the expected M. What could
 In (a) the past tense is formed by suffixation of [t]. This is unusual, in that a
occasion this seemingly bizarre innovation? past tense form in [d] would be expected. Compare quell quelled, compel,
The innovative plural resembles the plurals of 'Greek' nouns, such as theses compelled. For some verbs, such as spell, the 'irregular' spelt co-exists with the
(plural of thesis), hypotheses, analyses, parentheses, and several more, These plural regular spelled.
forms share several properties. Apart from the fact that they all end in [si:z] , they
consist of at least two syllables, often more, in which case word stress mostly falls three  In (b) nothing is added; rather, the final [d] is replaced by [t].
syllables from the end. The regular plural of process almost conforms with the  In (c) there is the suffixation of [t], as in (a). In addition, the stem vowel
plural schema for these nouns. The innovation consists in the plural being made to shortens from [id to [e].
fully conform with the schema. That process comes under the influence of the  In (d) there is once again suffixation of [t] along with the shortening of the
schema may well be due to the fact that words such as hypothesis, thesis, and others vowel to [e]. Since the base verb ends in a voiced consonant, suffixation of [d]
are associated with scholarly academic discourse; the innovative plural of process would be expected.
seems to be largely restricted to this context. As also noted in Chapter 1, the pattern  In (e) there is again vowel shortening in association with the suffixation of [t].
seems to be being extended to other plurals, such as biases and premises. These are In addition, the final stem consonant devoices: [v] 4 [t].
forms which, once again, tend to be restricted to academic contexts.
It would be possible to capture these five different processes by means of five different
The example of processes [prouszsitz ] suggests that the innovative plural has been rules, each performing a different operation on a different kind of input. Such an
formed, not by the application of a rule to an input form (prior to the creation of the approach would conceal the fact that the rules all conspirel4 to create a past tense
innovative form there was no such rule), but by the desire to make the plural conform form with a certain structure. The past tense forms are all, monosyllabic; the vowel
with a plural schema, a schema, moreover, which is probably quite well entrenched is in all cases a short [e]; and the past tenses all end in a voiceless consonant cluster,
for speakers familiar with academic discourse.'3 the second member of which is [t].'5 There is a sense, therefore, in which all the
verbs in (4) undergo the very same process: they are made to conform to the past
tense schema [ eCti.
12 Another example is provided by the word roof Mostly, the word forms its plural in the regular war
roofs, Occasionally, however, the form rooves is encountered; there are five examples in the BNC against
658 examples of roofs, while the COCA has 2,015 instances of roofs, with only one solitary example of
rooves. The irregular plural would appear to have been formed on the pattern provided by the dozen or so 14
On the notion of conspiracy, see Kisseberth (1970). The term refers to the fact that diverse rules,
nouns which exhibit fricative voicing in the plural. See Pierrehurnbert (2003) for a discussion of this
operating over different kinds of input, may all serve to create a common output.
example in terms of her notion of probabilistic phonology. 15
It is the voiceless final fti which occasions the change of iv] to (fi in left and cleft. According to a very
For languages with more complex morphology, such as German, the role of output schemas over
well-entrenched phonological construction in English, obstruent clusters must agree in voicing.
input rules is even more compelling; see K8pcke (1998).
136 Constructions Constructions and the autonomy of syntax 137

A striking example of the power of an output schema in contrast to source- instigates an event (consider: The rabbit was shot by the farmer or The next day saw
oriented rules is provided by the much-discussed way-construction (see e.g. Gold- our departure); nor is it possible to classify a word as a noun or a verb simply by
berg 1995; Jackendoff 1997). Here are some instances: referring to its semantics. It won't do to define 'noun' as the name of a person, place,
or thing, or to define 'verb' as the name of a state or action. Explosion, undeniably, is a
(5) a. I pushed my way to the exit. noun. We do not arrive at this judgement by contemplating the word's semantics.
b. I lied my way through the interview. Instead, we look at the syntactic behaviour of the word. The word distributes as a
c. The government hopes to spend its way out of recession. noun: it can take definite and indefinite determiners, it can pluralize, and it can be
d. We ate our way through France. (from a travel report) preceded by an adjective. If it behaves like a noun, then it is a noun, even though,
arguably, it has verb-like semantics.
The construction is associated with a rather specific semantics. The subject entity
moves (either literally or metaphorically) to a place designated by the PP, by V-ing. Considerations such as these easily lead to the view that syntax constitutes an
(There is also the nuance that some difficulty and effort are involved.) The construc- autonomous level of linguistic organization, mediating between, but independent of,
tion is remarkably productive, in the sense that a very wide range of verbs can both phonological and semantic structures. Such a view is inconsistent with the
feature in it. (Curiously, however, motion verbs such as go, walk, run, and drive Cognitive Grammar claim that language is basically and quintessentially symbolic,
tend not to be attracted to the construction.) Observe that the verbs appear in what consisting only of phonological and semantic representations and symbolic relations
looks like a transitive frame, more specifically, in a [_ NP PP) frame. Yet the between them.
construction easily tolerates verbs which are not subcategorized for such a frame. One way to counter the autonomy of syntax argument would be to address head-
Elsewhere in the language, lie (`speak untruthfully') is intransitive, while spend, on the claim that syntactic categories lack an identifying semantic content. This has
though transitive, cannot take way as its direct object (except in the way- been Langacker's strategy. Langacker (1987) strongly argued that noun and verb can
construction). It is the construction itself which sanctions this particular use of the be associated with a schematic semantic content, the former having to do with the
verbs. The demands of the construction override the properties of the participating status of the referent as a <thing', the latter designating a 'temporal relation'. is He has
verbs. also proposed schematic values for other elements of syntactic description, such as
In summary, evidence from within language itself, as well as from various experi- the subject and object relations, the subject of a clause being the more <prominent'
mental paradigms, strongly supports a construction-based account of linguistic entity in the relation designated by the verb. A problem with this account is that it
knowledge. Complex expressions are formed, not by rules operating over an input runs the danger of circularity. 'Thing' is defined so broadly (namely, as a 'region in a
string, but with reference to constructions, specified at various levels of abstraction. domain') that practically anything can be brought under its scope, the only criterion
Even regular processes, such as regular plural formation, can be understood in these for the correctness of the analysis being, precisely, the fact that concept in question is
terms. Before we rush to endorse a constructionist approach, however, there is an referred to by a nominal expression (Hudson 199o). In the last analysis, we determine
important issue that we need to tackle. This concerns the supposed autonomy of that a word is a noun because of its distribution, not because of its meaning, and we
syntax. If valid, the autonomy of syntax argument would undermine a central claim identify the subject of a clause on the basis of syntactic criteria, not by reference to its
of Cognitive Grammar and, with it, the basis for a constructionist account of prominence. This is not to dismiss out of hand Langacker's semantic characteriza-
syntax. I address these issues in the next section. tions. It could well be, however, that the semantic properties of nouns and subjects
that Langacker identified derive from their syntactic status, rather than their syntactic
status being the consequence of their semantics.
Constructions and the autonomy of syntax
Another approach was pursued by Croft (1991). Croft sought a functional account
As was mentioned at the beginning of this chapter, the Cognitive Grammar view of of the major categories of syntax, appealing to such notions as reference (a function
language departs significantly from mainstream and especially generative views in its
approach to syntax. On mainstream views, syntax constitutes an independent level of
organization, based on categories such as noun, verb, preposition, noun phrase, and 16 Taking a slightly different tack, Givan (1984) appealed to the semantic properties of prototypical
the like, as well as on relations between these elements, such as subject-of (a verb), nouns and verbs. Thus, the prototypical noun refers to a time-stable entity while a prototypical verb
refers to a transient or changing phenomenon. The trouble with this account is that while explosion may
direct object, and so on. Linguistics students learn very early in their career that these well be a non-prototypical noun in terms of its semantics, it displays all of the distributional properties of
categories and relations cannot be defined in semantic terms, even less, in phonolog- prototypical nouns. In terms of its syntactic behaviour, explosion is not at all a non-prototypical noun.
ical terms. You cannot identify the subject of a clause by searching for the actor which
3.38 Constructions Constructions and the autonomy of syntax 139

of nominals), predication (a verbal function), and modification (characteristic of likewise for all the other commonly recognized lexical and syntactic categories.
adjectival and adverbial elements). Deviations from this state of affairs—as when an Rather, we should be speaking of noun phrase and so on as a shorthand way of
adjectival or verbal form is nominalized, thus making it possible to predicate some referring to the set of items which are able to occupy a certain slot in a specific
property of the derived concept—are typically marked in some way, by special construction. In principle, then, there are as many noun phrase categories as there are
morphology, for example. Thus, explosion (a nominal) is marked vis-à-vis verbal constructions which have, as one of their parts, a noun phrase constituent. We
explode by means of the nominalizing -ion suffix. should have to recognize [Noun Phraseix—this being the noun phrase category
In more recent work, Croft (won 2007) has suggested a more radical approach, defined with respect to construction X—alongside [Noun Phrasely, the noun phrase
one which effectively turns the autonomy of syntax argument on its head. We do not category defined with respect to construction Y, and so on. To the extent that these
hesitate to classify explosion as a noun, neither are there any doubts about the verb noun phrase categories are coextensive there may be no harm in speaking of noun
status of explode. This is not because of what the words mean but because of the way phrases tout court, and similarly for all the other commonly recognized lexical and
they distribute in the language, that is, on the basis of their syntactic and morpho- syntactic categories. However, the possibility of construction-specific differences
logical properties. Now, to assign a word to a lexical category on distributional should not be overlooked.r
grounds is to appeal to the constructions in which the word can occur. If we say As an illustration, let us consider a couple of constructions which refer to noun
that a word is a verb because it can take a past tense inflection we have effectively phrases. One is the clausal construction [NP VP], where the subject slot is
defined the verb category with respect to the past tense construction. If the ability to occupied by a noun phrase, Another is the prenominal possessive construction [NP
form a noun phrase in association with the definite determiner is a feature of POSS N], where NP constitutes the possessor, POSS is the possessive morpheme
(common) nouns, we have defined (common) noun with respect to a noun phrase (typically 's or one of its variants), and N is a noun designating the possessed entity.
construction. By the same token, 'noun phrase' would be defined with respect to By and large, NPs which can occur in the transitive construction are also able to
function as possessors, and vice versa. The farmer is an NP in the clausal
those constructions which have noun phrases as a constituent, while clausal subjects
construction (The farmer shot a rabbit) as well as in the possessive construction
are defined with respect to clausal constructions, and so on.
(the farmer's wife). There are, however, some discrepancies. Mine can function as an
A phonological analogy illustrates the issues (Taylor 2004b). The distinction
NP in the clausal construction, but not in the possessive construction:
between vowels and consonants is fundamental to any phonological theory. We
might propose to define the two categories in terms of their inherent phonetic, that
is, their articulatory-acoustic properties, claiming that vowels are relatively sonorous (6) a. Mine was expensive.
elements, produced with relatively free flow of air through the oral cavity, whereas b. *Mine's price
consonants are produced with some kind of constriction of the air flow. We do not, Arguably, presentational there and meteorological it constitute the NP subject in
however, identify the sounds of a language as vowels or consonants simply by clausal constructions. Thus, in interrogative contexts, these items invert with
inspecting their phonetic properties; we refer to their role within a larger construc- respect to an inflection-bearing auxiliary verb, this being a typical property of a
tion, namely, the syllable. (In fact, were it not for the role of the syllable in clausal subject. However, presentational there and meteorological it are excluded
phonological structure, there would be no point in making a distinction between from featuring as possessors:
consonants and vowels in the first place.) Acoustically, the initial sound of yes is very
similar to the final sound of say. Yet we would regard the initial [j] of yes as a (7) a. There is a man at the door e•-, Is there a man at the door?
consonant and the [j] off-glide of say as a vowel. This is not because of their inherent b. *There's man
acoustic-articulatory properties but because of their role within the syllable con- (8) a. It was raining e- Was it raining?
struction. The [j] of yes occupies the consonant slot of the syllable construction; b. *Its weather
the [j) of say occupies a vowel slot. While no doubt grounded on facts of articulation,
membership in the categories is defined, in the last analysis, in terms of the (phono-
logical) constructions in which they occur.
Seen in this light, constructions are primary and syntactic and lexical categories are
17
Croft's approach entails that syntactic categories are not only construction-specific but also language-
derivative on the constructions in which they are eligible to occur. This approach specific. There may be no basis, other than terminological convenience, to refer to nouns, noun phrases,
suggests that it may be an error to speak of nouns or noun phrases as such, and and so on, as universal categories, applicable to all languages.
1
140 Constructions Collostructional analysis 41

The upshot is that the noun phrase category defined with respect to the possessive collocate with the construction, that is, which are able to occur in the open slot
construction is not the same as the noun phrase category defined with respect to the positions. By taking account of the number of potential candidates available to fill a
clause construction." slot, and the overall frequency of these candidates in the construction and in the
language at large, it is possible to estimate the degree of attraction (or,
conversely, repulsion) between an item and the construction.19
Collostructional analysis
To illustrate, let us consider a couple of Stefanowitsch and Gries's examples. The
A basic assumption of the dictionary plus grammar book model is that all first concerns the past tense construction, schematically [V + past tense]. All verbs—
members of a category behave identically with respect to the rules which refer to the with the exception of some of the modals and a few other defective verbs (see
category. All items marked in the lexicon as a common noun are equally available to p. 6o)—.can occur in the construction.2° Yet not all verbs are equally likely to occur
feature in the noun phrase structure [DET N]; all noun phrases can function as the in the past tense. Some verbs, such as say, become, and tell, are biased towards
occurring in the past tense vis-à-vis the norm for verbs as a whole; others, such as
subject or direct object of a transitive verb or as the complement of a preposition; all
hope, remember, and work, are biased towards occurring in the present tense.
transitive verbs are equally eligible to appear in a passive configuration, and so on.
Another of their examples concerns the waiting to happen construction:
Indeed, it is these very equivalences which justify the recognition of the lexical arid
syntactic categories in the first place. Explosion is regarded as a noun precisely
(9) (It was) [a N] waiting to happen.
because it distributes in the language like other members of the category.
The discussion in the preceding section suggests a radically different view of the In principle, any event-denoting noun can occur in the construction. Examples (from
relation between the lexicon and the syntax, between words and the configurations in the COCA) include crisis, tragedy, scandal, injury, nightmare, revolution, riot, over-
which they occur. On the one hand, a construction specifies the kinds of words which dose, crime, bad ending, and another Enron. It will be noted that all these nominals
are eligible to occur in it. At the same time, words are specified for the range of refer to lad' events; the construction thus exhibits a negative prosody (p. no). One
constructions in which they are eligible to occur. Thus, to take up one of our earlier can, of course, imagine 'good' things appearing in the construction: It was a marriage
examples (see pp. 51-3), lap occurs preferentially in a possessive environment, waiting to happen. Such expressions, however, tend not to be attested. Not only is the
while the possessive expression itself—my lap, the lap of luxury, and so on— construction associated with 'bad' events, the two most common words to occur
occurs preferentially as the complement of a preposition, typically in or on. in the construction are accident and disaster. The BNC contains twenty-six tokens
Again with reference to an earlier example (see p. 55), the noun slot in the of the construction, exemplifying nine different nouns. Of these, accident and
[THERE'S NO N IN V-mo] construction is occupied, preferentially, by fun, disaster each occurs nine times, together making up about two-thirds of all
point, use, harm, and sense, while the noun slot in the [IT'S NO N V-ING] tokens.
construction is occupied, preferentially, by fun, good, use, joke, and big deal. The
two sets of nouns only partially overlap. An important aspect of knowing a word By shifting the perspective of the analysis we can focus on a word or some other
such as fun is to know just which constructions it is able to occur in. linguistic unit and consider the kinds of construction in which it preferentially
occurs. While the [wAITING TO HAPPEN] construction is very strongly associated with
The interaction of linguistic expressions and the contexts in which they occur has the words accident and disaster, these two words show no particular association with
been insightfully studied by Gries and Stefanowitsch in terms of couosTRucnoms the construction. There are 6,30o tokens of accident and 2,770 tokens of disaster
(Stefanowitsch and Gries 2003; see also Gries, Hampe, and Schornefeld 2005; Gries in the BNC; only a minuscule proportion of these occur in the waiting to happen
and Stefanowitsch 2004). (The term itself is a blend of collocation and construction.) environment.2'
The starting point for a collostructional analysis is a construction, understood as an
internally complex syntactic configuration containing one or more open slots. 19
Strictly speaking, collostructional analysis is based, not on the notion of collocation, but on the
The concept of colligation. Collocation, it will be recalled (see p,106), concerns the degree of attraction between
words, while colligation (see p. 109) has to do with the preferred syntactic environment of an item.

Whether a verb forms its past tense through regular suffixation of {t} or by other means is not
relevant to the discussion.
See Chapter 4, footnote a, for a further example of a word category defined with respect to a
18
21
In spite of this, it could be argued that disaster is somewhat more closely associated with the
construction. At issue is the set of items which are able to occur as 'particles' in so-called phrasal verbs,
construction than accident. Overall, disaster is a less frequent word than accident. This being so, a slightly
such as take up, bring in, set out, put aside, take apart, etc. In the main, the particles are members of the
greater percentage of all instances of disaster occur in the [WAITING TO HAPPEN] construction than is the case
category of preposition. Yet the categories of preposition and verb particle are not coextensive. Aside and
with accident.
apart, for example, are not prepositions. What counts as a particle with respect to phrasal verbs is
determined by the set of phrasal verbs. analysis aims to identify those items which
142 Constructions Constructions all the way up? 143

Some words are very choosy with regard to the constructions in which they which fits the specification of the rule. Thus, once a child starts producing passive
occur.22 Examples already discussed include lap and whereabouts. Multi word clauses—evidence that she has acquired the passive rule—one might suppose that
phrases, too, can be highly selective. The double negative not for nothing occurs every transitive verb in the learner's lexicon will be available for use in passive clauses.
mainly in two environments. The favoured environment is as the initial element of This, however, is not how acquisition proceeds (Tomasello 2003, 2006). Construc-
the negative inversion construction; this environment accounts for about So per tions are acquired on a piecemeal fashion, verb by verb. Passive clauses, for example,
cent of all instances in the BNC. are associated, in the first instance, only with specific verbs (Brooks and Tomasello
1999). Initially, therefore, a passive clause has the status of a verb-specific idiom. It is
(to) a. Not for nothing was he known as "the Destroyer".
only at a later stage, when a critical number of such idioms have been learned, that
b. Not for nothing are black holes called black. generalization to new verbs occurs. The same goes for the structure of noun phrases,
c. Not for nothing have the invitations dwindled a bit over the years.
and indeed for many other constructions. One might suppose that once the learner
The remaining instances occur in a focusing construction introduced by it is/was:23 starts using nouns in a range of NP constructions, with a range of determiners such
as the, a, this, my, and so on, every eligible noun will appear with the full range of
(22) a. It is not for nothing that scabies is commonly known as the itch.
determiners. It turns out, however, that initially certain nouns are closely associated
b. It was not for nothing that he was chosen as Mr Squeaky Clean after the
with particular determiners (Pine and Lieven 3997). It is only at a later stage, when a
sexual and financial aberrations of his two predecessors.
critical mass of lexically specified constructions has been acquired, that the learner
Neither of these constructions shows a particularly strong association with not for proceeds to the appropriate generalization, namely, that any (common) noun can be
nothing., practically any negative expression can introduce the negative inversion used with the full range of determiners. Even so, as our case study of lap and bosom
construction, while a wide range of items are able to occur in the focusing construction. showed, distributional preferences may persist, even into the adult language.
In the examples discussed, there are notable asymmetries in the relation between a
construction and the items that can occur in it. In some cases, a construction is very
Constructions all the way up?
strongly associated with a particular word, or small set of words; in other cases, a
word or larger expression is strongly attracted to a certain construction, or small set In discussing constructional idioms of the kind Off with his head, I stated that these
of constructions. Sometimes, however, the relation between word and construction is expressions were idiomatic to the extent that their properties could not be derived
more evenly balanced. A very large number of different verbs are able to occur in the from the general syntactic rules of the language (see p. 40).
ditransitive [V NP NP] construction. However, the verb which is most strongly This formulation suggests that alongside idiomatic constructions there exist other
attracted to the construction, and the one which the construction most strongly constructions which are not at all idiomatic and which can be generated from the
attracts, is the verb give (Goldberg 2006; Gries and Stefanowitsch 2004). Goldberg has general combinatory rules of the syntax. The implication is that the expressions in a
argued that this strong mutual attraction between a verb and a construction con- language can be partitioned into two sets. One set comprises expressions which are in
some way idiomatic, idiosyncratic, or exceptional; the other set comprises perfectly
tributes significantly to the learnability of a construction, in both its formal and
regular expressions which conform to general syntactic rules. Whereas Off with his
semantic aspects. This is a topic which I will take up again in Chapter 8.
head needs to be described in terms of a low-level construction, involving, amongst
other things, the construction-specific use of the word with, other expressions, such
Acquisition as The farmer shot a rabbit, can be generated by the syntax operating over the lexicon.

Given the assumptions of the generative model, one might suppose that once the The discussion in this chapter suggests that this distinction may not be valid. The
learner has acquired a syntactic rule, the rule will apply across the board to every item (OFF WITH] construction and the transitive clause construction differ only with respect
to their generality, not in terms of their status as constructions. The former, as we
have seen, imposes tight restrictions on the kinds of items that can occur in it; it is
Cranberry words (see p. 71), by definition, occur only in a specific environment. Intents is restricted to
22 also associated with a specific semantics. The transitive clause construction, on the
the context to all intents (and purposes), once-over occurs in the context give (someone or something) the other hand, is relatively unconstrained, both with respect to the items which can
once-over, and so on. feature in it and with respect to its semantics.
It is also worth noting that the majority of examples of not for nothing emphasize the appropriateness
23

of the way in which something has been named.


144 Constructions Constructions all the way up? 145

The crucial word here is 'relatively'. As a matter of fact, the transitive clause In brief, even the most general syntactic patterns of a language, concerning the
construction is not entirely free of idiosyncratic aspects. The idiosyncrasies are major sentence types, need to be regarded as constructions. Their perceived
language-specific and therefore have to be learned (Taylor 2003b: Chapter 12). generality is due to the fact that a very wide range of items can occupy their various
Consider the range of verbs which are eligible to occur in the construction. These slots; semantically, too, the constructions are compatible with a range of possibilities.
are commonly referred to as transitive verbs. While a semantic characterization of Nevertheless, the constructions are not without their idiosyncratic properties,
(some) transitive verbs is certainly possible—we would probably want to say that the which will need to be specifically learned. Not only this, but the very fact that a
verbs designate some action performed by an actor which impinges on, and affects, construction exists in a language at all is something which needs to be learned. The
another entity, a patient—what counts as a transitive verb is ultimately decided by the idea that constructions are no more than the output of rules operating over the
possibility of the verb occurring in the construction. In English, remember counts as lexicon cannot be maintained. It's constructions all the way up.
transitive: I remember that day. The French and German translation equivalents of
remember are not transitive: le me souviens de ce jour, Ich erinnere mich an den Tag.
In both languages, the verbs are reflexive, with the English direct object appearing
in a prepositional phrase.
The notion becomes even more compelling when we consider some other major
sentence types, such as the ditransitive (V NP NP], as in give the dog a bone. Such a
construction simply does not exist in the Romance languages. In French, Italian, and
so on, the goal, or recipient, must appear in a prepositional phrase: 'give a bone to the
dog'. For those languages which do have a ditransitive construction, such as Korean
or the Bantu languages, the range of application of the construction may differ very
considerably from that of English (Jackendoff 1996). Consider the following Zulu
examples (Taylor 1997):

(12) a. Umama unika amantombazana imali.


Mother gives (the) girls money
b. Umama utschela amantombazana indaba.
Mother tells (the) girls (the) news
(13) a. Bayikhuthuza indoda isilchwamasayo.
They robbed (the) man his wallet
`They robbed the man of his wallet'
b. Udokotela ukhipe umfana izinyo.
The doctor extracted (the) boy (the) tooth
`The doctor extracted the boy's tooth'
c. Angimazi lomuntu igama.
I do not know this person (the) name
do not know this person's name'
The examples in (12) correspond to their English equivalents. In (13), however, we
see that the ditransitive construction is used where English would have a different
wording. In (13a) the postverbal nominal refers to a person who is deprived of
something, in (13b) the nominal refers to a person who is affected by virtue of a process
directed at a body part, while in (13c) the nominal indicates a possessive relation.

Vous aimerez peut-être aussi