Scarica Historical linguistics e più Schemi e mappe concettuali in PDF di Lingue solo su Docsity!
Theoretical contexts for the
study of lexicalization and
grammaticalization
1.0 Purpose of the present study
(1) We are celebrating a fascinating holiday today.
is something we might well say to a visitor from abroad, and not think twice
about whether holiday and today or the -ing of celebrating and of fascinating
function differently from the point of view of our knowledge of language.
However, linguists, grammarians, and others who study and think about
language, how it is structured, how we come to know it, and how it changes
concern themselves with just such questions. In the introduction to the
Concise Oxford Dictionary of Linguistics, Matthews says: ‘‘Everyone will
agree that linguistics is concerned with the lexical and grammatical cate-
gories of individual languages’’ ( 1997 :vi), and this is what our example in (1)
is about: holiday, celebrate, and fascinating are usually regarded as ‘‘lexical,’’
members of large, ‘‘open’’ classes of forms that are relatively infrequently
used and express relatively concrete meaning, while we, are, and a are
regarded as ‘‘grammatical,’’ members of smaller, relatively ‘‘closed’’ classes
of forms that are very frequently used and express relatively abstract mean-
ing. Moreover, today is not clearly a lexical or a grammatical form, having
partially concrete and partially abstract meaning, and belonging to a
rather large set of adverbs. Finally, the -ing of celebrating and the -ing of
fascinating, although seen as originating in the same grammatical form,
are generally understood as having developed differently over time, the
former remaining grammatical and the latter becoming lexical. What these
differences mean, how this kind of distinction plays out in language
change, and what research questions it suggests are among the topics of the
present book.
1
In recent years questions have frequently been raised about the relation-
ship between ‘‘lexicalization’’ and ‘‘grammaticalization.’’ The two terms, like
many other linguistic terms, have been used to refer ambiguously to phe-
nomena viewed from the perspectives of relative stasis (‘‘synchrony’’) or of
change over time (‘‘diachrony’’), to the process and to the results of the
process, and also to theoretical constructs modeling these phenomena.
According to Lehmann ( 1995 [1982]:6), the first formulation of an
opposition between lexicalization and grammaticalization was Jakobson’s
( 1971 [1959]) characterization of the first as optional, the second as obliga-
tory. Since then, they have been theorized in a number of different ways,
sometimes totally independently of each other, sometimes together. One
constant in all these uses is pairing of meaning and form, and the extent to
which this pairing is systematic or idiosyncratic. The starting point of the
present work is to bring together a variety of scholarly debates concerning
this relationship in language change, with focus on lexicalization, which has
been studied far less systematically than grammaticalization.
The first three chapters are reviews of the literature; the last three propose
some solutions. In this chapter, we will briefly introduce the contexts for
the study of lexicalization and grammaticalization, most especially on
approaches to grammar, lexicon, language change, lexicalization, and
grammaticalization. We will not attempt to resolve the differences of opinion.
Chapter 2 focuses in more detail on lexicalization, especially the definitions
and viewpoints that have emerged during the last fifty years of work in
linguistics. Chapter 3 presents recent arguments concerning the similarities
and differences between lexicalization and grammaticalization. Chapter 4
suggests one possible integrated approach to lexicalization and grammatica-
lization that resolves the major debates about their relationship. Chapter 5
addresses some particular problems in the history of English from the
perspective of definitions developed in Chapter 4 , and Chapter 6 summarizes
the book, ending with suggestions for further directions for research.
1.1 Debates concerning grammar and language change
It is impossible to understand how either lexicalization or grammaticalization
have been conceptualized without paying attention to underlying assump-
tions about grammar and its relationship to the lexicon, as well as underlying
assumptions concerning the dynamics of language change. A full investiga-
tion of these topics would entail a detailed history of linguistics, especially in
the twentieth century. Space allows only for some sweeping generalizations
here, which, unfortunately, tend to polarize and to be caricatures. However,
without some attention to different foundational assumptions, it is often
difficult to make sense of the literature or to propose a possible solution to
the many issues that have been raised. In Section 1.1.1 we summarize two
extreme approaches to grammar, the polar opposites between which much
2 Theoretical contexts
how grammars may vary cross-linguistically. Universals of language are
considered to be tendencies, not absolutes, and are usually of a general
cognitive nature, not autonomous and not specific to language.
The turn of the present century has seen the emergence of several
possibilities for a meeting of minds, as some generative linguists begin to
try to account for cognition-based structures (e.g., Jackendoff 1983 , 2002 ),
for productivity (e.g., Jackendoff 2002 ), for the dynamic, emergent proper-
ties of the speaker’s knowledge of the system (e.g., Culicover and Nowak
2003 ), and for the variation that undeniably occurs in language (see work on
Optimality Theory, e.g., Boersma and Hayes 2001 ; Lee 2001 ; Bresnan,
Dingare, and Manning 2002 ). Moreover, some ‘‘functional’’ linguists have
sought to formalize their work at least in part (see, e.g., Bybee and Hopper
2001 for frequency studies; Croft 2001 for syntax).
Common to many, but by no means all, theories is the notion of ‘‘grammar’’
(whether at the abstract level of Universal Grammar, or at the more empirical
level of the grammar of a particular language) that is distinct from the notion of
‘‘lexicon.’’ If such a distinction is made, ‘‘grammar’’ is the set of categories,
patterns, and organizing principles evidenced by language, most essentially
abstract patterns of semantics, syntax, morphology, and phonology that at
least in theory permit infinite combinations. By contrast, the ‘‘lexicon’’ is a finite
list (for any individual) of (more-or-less) fixed structural elements that may be
combined. The lexicon is typically a theoretical concept, as distinguished from
a ‘‘dictionary,’’ which is a practical description. Hence, there is discussion of a
‘‘mental lexicon’’ (an abstraction and idealization), not of a ‘‘mental dictionary’’
(Matthews 1997 :s.v. ‘‘lexicon’’).
2
There have been essentially two views of the relationship of the lexicon to
the grammar in generative theory of the last fifty years. The first, which
Jackendoff ( 2002 ) calls the ‘‘syntacticocentric approach,’’ assumes that the
lexicon is a list of idiosyncratic items which are selected and inserted into
syntactic structures (see various versions of generative syntax from
‘‘Standard Theory’’ [Chomsky 1965 ] through the Minimalist Program
[Chomsky 1995 ]). Phonological and semantic interpretations are derived
from the lexicon together with the syntax. The second, proposed by
Jackendoff ( 1997 , 2002 ), provides an alternative architecture: one in which
phonological, syntactic, and conceptual structures are parallel components
of the faculty of language, and in which lexical items ‘‘establish the corre-
spondence of certain syntactic constituents with phonological and concep-
tual structures’’ (Jackendoff 2002 :131).
3
A key proposal in Jackendoff ’s
2
As we will see, while some theories of the lexicon are roughly equivalent to ‘‘voca-
bulary,’’ many are not, since they include grammatical forms such as past tense -d.
3
Other proposals that treat the lexicon as part of the combinatorial architecture of a
complex set of parallel structures include Lexical Functional Grammar (e.g.,
Bresnan 2001 ), and various types of Head Driven Phrase Structure Grammar
4 Theoretical contexts
work is that the lexicon is multistructured and includes not only highly
idiosyncratic, but also more regular elements. This is more in keeping with
many functionalist views of the lexicon, which point to parallels between
lexical and grammatical organization, although the regularities may be
considered to belong to morphology rather than the lexicon (see, e.g.,
Bybee 1985 , 1988 ; Langacker 1987 ; Haspelmath 2002 ). A more detailed
discussion of Jackendoff ’s views, with focus on the problem of distinguish-
ing types of lexical categories, follows in Section 1.2.
1.1.2 Approaches to language change
Because lexicalization (and grammaticalization) will here be conceptualized
primarily as historical processes subject to normal constraints on language
change,
4
we will briefly set out some assumptions concerning language
change before turning to a more detailed examination of the conception of
the lexicon. While a comprehensive examination of theories of language
change is far beyond the scope of this introduction,
5
we will mention here a
few factors that will help illuminate the debates over lexicalization and
grammaticalization.
Historical linguistics was the focal point of attention in the nineteenth
century, during which time many foundational ideas of linguistics were
developed, most especially the concepts of structure and pattern.
Discovery of such sound laws as Grimm’s Law, which showed how the
Germanic languages differed systematically in consonant articulation from
the other Indo-European languages, and the Great Vowel Shift, which
showed how later English differed systematically from earlier English with
respect to the place of articulation of the long (later tense) vowels, high-
lighted the ways in which language phenomena are structured.
6
Work on
(e.g., Pollard and Sag 1994 ) and Construction Grammar (e.g., Goldberg 1995 ;
Fillmore, Kay, Michaelis, and Sag 2003 ).
4
There has recently been some confusion about the term ‘‘process,’’ depending on
whether the term is used restrictively or not. As Newmeyer rightly points out,
grammaticalization is not a ‘‘distinct process’’ in the restrictive sense that it is ‘‘an
encapsulated phenomenon, governed by its own set of laws’’ ( 1998 :234). He also
acknowledges that ‘‘process’’ is most usually used in a non-restrictive sense as a
‘‘phenomenon to be explained’’ (232). We use the term in a slightly different but
even more usual non-restrictive way to focus attention on (a) the need for a
dynamic perspective, (b) the micro-steps that are obscured by the ‘‘>’’ typical of
representations of change.
5
For the state of historical linguistics at the end of the twentieth century, see Joseph
and Janda ( 2003 ).
6
The term ‘‘Great Vowel Shift’’ appears to be attributable to Otto Jespersen, see
Jespersen ( 1961 [1909–1941]:Vol. I, Chap. 8).
Debates concerning grammar and language change 5
(b) The transition problem: What are the intervening stages that define the
path by which A gives rise to B (and typically coexists with it for at least a
while)? Change proceeds by small steps, not large leaps (although accu-
mulations of changes may have cascade-like effects that lead to more
substantial change). Change by small steps will be discussed more fully
under ‘‘gradualness’’ in Section 1.4.2.
(c) The actuation problem: How does change start, when and where does it
start (‘‘actuation’’) and how does it spread through the system
(‘‘actualization’’)?
One type of change that is particularly important for our discussion of
lexicalization and grammaticalization is ‘‘reanalysis.’’ In a foundational
paper on the topic of syntactic reanalysis, Langacker defined it as covert
change: ‘‘change in the structure of an expression or class of expressions that
does not involve any immediate or intrinsic modification of its surface
manifestation’’ ( 1977 :58). From this perspective reanalysis involves:
(a) change in constituency, or what goes with what (e.g., change in mor-
phological bracketing of [a] napron > [an] apron
7
(b) a change in category labels (e.g., main verb > auxiliary),
(c) boundary loss (e.g., be going to > gonna).
Reanalysis is not restricted to morphosyntax: when a lexeme develops new
polysemies (e.g., silly ‘blessed, innocent’ > ‘foolish’), it has undergone
semantic reanalysis.
Another major type of change is analogy: the generalization of a struc-
ture (see Kiparsky 1992 ). By contrast to reanalysis, analogy is overt, and
indeed it is often only through analogy that reanalysis can be detected (see
Timberlake 1977 ). Thus, as we will see, when the motion verb construction
be going to is reanalyzed as a future auxiliary, evidence for this reanalysis
comes from use of be going to with verbs that do not ordinarily collocate
with motion, e.g., verbs of psychological experience such as like or know.
8
Weinreich, Labov, and Herzog ( 1968 ) made a key distinction between
innovation (changes that happen in the individual) and change (changes
that spread to others) (see also Milroy 1992 ; and discussion in Janda and
Joseph 2003 ). In doing so, they proposed a significantly more social view of
change than the generative view with which it competed (e.g., Lightfoot
1979 ). On the generative view, change is equivalent to innovation and is to
be found in differences between cognitive states of individuals. Since the
focus is on internalized systems, the interest in change lies in how grammars,
i.e., internalized sets of patterns and relationships among patterns, change
7
Naperon ‘small tablecloth’ is a borrowing from OFr.
8
Detailed discussion of reanalysis and analogy can be found in Harris and
Campbell ( 1995 ).
Debates concerning grammar and language change 7
(see, e.g., Kiparsky 1968 ; Kroch 2001 ). Syntax is privileged as central and
autonomous, not at all or only marginally affected by either semantics or
phonology, and the child is privileged as the locus of change, since the small
child has to learn the language from scratch.
Toward the end of the twentieth century a dialog developed as ‘‘function-
alist’’ theorists sought to greater and lesser extents to integrate the advances in
formal linguistics with a perspective on language that paid more attention to
construal of meaning and to dynamic aspects of language (e.g., Bybee,
Perkins, and Pagliuca 1994 ; Croft 2000 ). This was often combined with
cross-linguistic work on language typology spearheaded by Greenberg (see,
e.g., Greenberg, Ferguson, and Moravcsik 1978 ). Furthermore, there was a
shift away from focus on strictly ‘‘internal’’ change to concern for the role of
the speaker in the ‘‘basic’’ two-person interactional dyad, and in the commu-
nity (see especially Milroy 2003 ). On this view, not only the child as hearer, but
also the adult (especially the young adult) as producer, can be the innovator
and therefore the catalyst for change (see, e.g., Haspelmath 1999a). We speak
of ‘‘language change,’’ yet strictly speaking, this is a misnomer – it is not
language in general or a language in particular that changes; rather, commu-
nities of speakers develop different representations of a system. Much of the
work on grammaticalization developed in this theoretical context. Since this
book is about lexicalization and grammaticalization as types of linguistic
change, our approach is largely functional-typological in orientation.
9
For historical linguists of a functionalist persuasion, the object of study
is how language systems can change over time, as attested by written textual
data (also spoken data since audio-recordings have become available).
A major contributing factor in the growth of historical linguistics at the
end of the twentieth century has been the advent of computerized corpora,
which give easy access to information about linguistic contexts for change,
frequency, and other factors important in answering the questions posed by
Weinrich, Labov and Herzog. For historical work on English the Helsinki
Corpus has been a major source of data (see Rissanen, Kyto ¨
, and Palander-
Collin 1993 ).
10
Such corpora for the most part reflect not the language of
small children, but the rhetorical practices and strategic interactions of
speakers with hearers, and have suggested to many that a theory of language
change needs to be usage- or ‘‘utterance’’-based, paying attention to mean-
ing and discourse function (Hopper and Traugott 1993 , 2003 ; Croft 2000 ;
Traugott and Dasher 2002 ). Furthermore, corpora have confirmed that
9
For possibilities of combining strictly formal with functional approaches in
historical study, see, e.g., Clark ( 2004 ).
10
Kyto¨ ( 1996 ) provides a key to the data in the Helsinki Corpus. This corpus and
several other corpora of English written and spoken language are available in the
International Computer Archives of Modern English (ICAME 1999).
8 Theoretical contexts
represent properties of the world (reference), but rather innate properties of
the mind that determine the way in which the world is conceived. For
example, Fillmore says: ‘‘the ultimate terms of a semantic description I
take to be such presumably biologically given notions as identity, time,
space, body movement, territory, fear, etc.’’ ( 1970 :111).
11
Such semantic
components reflect systematic relations that hold among items in the voca-
bulary of languages and can be used to compare cross-linguistic correlations
between meaning and form, known as ‘‘lexicalization patterns,’’ and to
make generalizations about constraints on these correlations. For example,
in the case of a ‘‘lexical field,’’ such as verbs of perception, Viberg ( 1983 )
proposes that there is a lexicalization hierarchy on two dimensions. One has
to do with the number of lexical distinctions available for the verbs of the
senses. For example, English distinguishes agentive and experiential look at,
see or listen, hear, but other languages like Hindi do not, and have only one
verb for look at/see and another for listen/hear. There are no languages with
one verb for look at/see, and two separate verbs for listen and hear. On the
second dimension, which concerns verbal complexity, ‘‘[i]f look at (or in a
few cases see) is expressed by a morphologically complex form, so is listen to
(or hear). But the opposite does not necessarily hold’’ (Viberg 1983 :136).
Such theories of lexical components and most especially more recent the-
ories about the way they combine to form lexical representations of complex
meanings play an integral role in discussions of ‘‘synchronic lexicalization’’
(see 1.3.1 below). They assume that there are universal semantic
components – which being universal are not learned – and various lan-
guage-specific combinatorial possibilities.
Note that a different term, LEXEME, is also sometimes used. While a
lexical item may be understood as any member of the lexicon (whether
primarily lexical or primarily grammatical), a LEXEME is typically contrasted
with a grammatical morpheme, or GRAM such as PL (¼ plural) (Bybee,
Perkins, and Pagliuca 1994 ). A lexeme refers to a word considered as an
abstraction such as RUN rather than as its various concrete WORD FORMS, or
grammatical modifications, such as run, runs, ran, running, or FOOT rather
than foot, foot’s, feet (Matthews 1997 ; Haspelmath 2002 ). From this per-
spective, the lexeme FOOT is a ‘‘stem,’’ and plural (the vowel alternation in
this case) is a grammatical affix.
11
See also, with various proposals, Katz and Fodor ( 1963 ), McCawley ( 1968 ),
Bierwisch ( 1970 ), Voyles ( 1973 ), Leech ( 1981 [1974]), Gruber ( 1976 ), and Lyons
( 1977 ). Initial theories about components of meaning have been developed more
fully in the light of questions about interfaces between meaning and syntax in,
e.g., McCawley ( 1968 ), Jackendoff ( 1983 , 2002 ), Talmy ( 1985 , 2000 ), Wierzbicka
( 1985 ), Dowty ( 1991 ), Levin ( 1993 ), Levin and Rappaport Hovav ( 1995 ),
Pustejovsky ( 1995 ), among others.
10 Theoretical contexts
There are different ways of conceptualizing the relationship between
units in the lexicon and the combinatorial possibilities they allow. One
view is that there are basically two mechanisms in the mind, two ways of
knowing: a ‘‘memory system’’ that stores and retrieves individual words, and
a ‘‘system of symbolic computation’’ that generates grammatical combin-
ations of words. More specifically, the lexicon, which stores irregularities
such as eat–ate, foot–feet, is distinct from the grammar, which provides a set
of regular syntactic and phonological rules of combination such as we find
in walk–walked, dog–dogs (Pinker 1999 ). Here lexical and grammatical units
can be sharply distinguished (according to regularity rather than function,
however). Others point out that not all combinatorial possibilities are
regular. Some combinations, especially those that have to do with the
make-up of the lexicon, are so specific to individual items and constructions
that they should be accounted for in the lexicon in terms of ‘‘lexical rules.’’
For example, take (someone) to task is relatively frozen compared to take
someone to New York and allows a passive, but not substitution of to task by
other noun phrases; the past tense of bring–brought has only a partial
phonological connection with the regular past tense -d. Anything stored in
long-term memory (including forms like lexical take to task and grammat-
ical -d) should be considered a ‘‘lexical item’’ (see, e.g., Sag and Pollard 1991 ;
and, for a considerably different view of the lexicon, Jackendoff 2002 ).
1.2.2 Categories of the lexicon: distinctions between lexical
and grammatical categories
As we have seen, the formal units of the lexicon are lexical items.
Semantically, the lexicon is said to express ‘‘lexical meaning.’’ The adjective
LEXICAL has a wider meaning than the nominal terms ‘‘lexical item,’’ ‘‘lex-
eme,’’ or ‘‘lexicon.’’ Lehmann observes that ‘‘lexical’’ can mean ‘‘1) belonging
to the inventory, 2) having a specific, concrete meaning’’ ( 2002 :14). The
adjective GRAMMATICAL likewise has wide meaning; it can refer to (1) con-
forming to the rules of the grammar, (2) having an abstract, structural/
functional, or indexical meaning.
12
There are a number of technical con-
cepts in linguistics that highlight these rather different meanings. For
12
Indexical meanings are those that point to aspects of context. Such contexts may
be external to the speaking event (e.g., that in Jill wants that) or internal to the
utterance (it in Jill dropped the pen and picked it up). They may also involve
speaker attitude, such as assessment of where some contentful item should be
placed on a value scale (focus particles, e.g., even in Even the NGOs left), assess-
ment whether a proposition refers to an event prior to, simultaneous with, or
expected to be later than time of utterance (tense), or indication of whether a term
is intended to be understood as new/non-recoverable or recoverable in the dis-
course (articles, cf. a boy, the boy). A useful term is ‘‘procedural’’ (see Blakemore
Concepts of the lexicon 11
marker such as the auxiliary will (in I will be giving a reading from my novel,
where futurity is represented explicitly by will, but is referentially inexplicit).
In grappling with the problem of parts of speech, much recent generative
work has focused on syntactic definitions. Four major universal categories
are usually posited: (N)oun, (V)erb, (Adj)ective, and (Prep)osition.
13
These
categories are called ‘‘lexical’’ because they dominate lexical items that are
semantically ‘‘whole,’’ but they are conceptually syntactic features. N, V,
Adj, and Prep can enter into complex phrases (NP, VP, AdjP, PrepP). The
main motivation for this proposal is to demonstrate the similarities in
phrasal structures involving the four categories; for example, according to
early versions of the theory, each may be ‘‘specified’’ by a functional cate-
gory such as determiner, auxiliary, or intensifier (the for many nouns, very
for many adjectives, right for many prepositions, e.g., right in back of the
bus). The four ‘‘lexical categories’’ contrast with ‘‘functional categories’’ like
determiners that are not semantically rich, are stressless under most circum-
stances, and are members of a minor, closed class (see below) (Napoli 1993 ).
However, few criteria are given to determine how to classify particular
items in particular languages as one or other of these parts of speech (Croft
2001 ). Furthermore, as has often been pointed out, the behavior of these
lexical categories may differ from one language family to another. For
example, Adjs are treated like Ns in many European, North African, and
Australian languages, but like Vs in the languages of North America, East
and Southeast Asia (see, e.g., Dixon 1982 ; Lehmann 1990 discusses the
structural consequences for the grammar of absence of the category Adj in
a language). In addition, the four major categories are often not specifiable
for a particular language. This has led to challenges from a number of
functional-typological linguists, some of whom suggest that labels like N,
V, Adj, and Prep are categories of particular languages, but are not uni-
versal. For example, Bhat ( 2000 ) suggests that languages can be distin-
guished according to the number of open classes of categories that have
independent morphological characteristics, or are ‘‘lexicalized’’: three
(nouns, verbs, adjectives) or two (nouns and verbs); in the case of the latter,
further distinctions can be made according to whether adjectives pattern
with nouns or with verbs. Even the universality of N vs. V has been
challenged; for example, Sasse ( 1988 ) questions whether Iroquoian lan-
guages have nouns on the grounds that supposed Ns take a prefix similar
to one found on verbs and may have temporal properties such as punctual.
This particular question has been answered in detail in Mithun ( 2000 ),
where it is shown that there are clear formal differences between Ns and
Vs in Iroquoian languages. However, because morphological Vs may be
13
Prep is more appropriately understood as ‘‘Adposition,’’ since whether this is a
preposition or postposition depends on the word order of the language in
question.
Concepts of the lexicon 13
used as names for entities, they function as Ns, e.g., akya:t:"(:to( hk ‘we are
related/my cousin.’ Although many morphological verbs have been lexi-
calized, i.e., frozen as Ns, the N–V distinction remains robust (Mithun
The functional-typological solution to the problem of parts of speech
focuses not on syntactic but on cognitive, conceptual structures.
14
In this
tradition, only N, V, and Adj (not Prep) are considered to be fundamental
to linguistic structure. Langacker ( 1987 ) takes the position that although some
basic cognitive structures are universal, grammars, including semantic struc-
tures, are language-specific. Language is a symbolic structure with various
interrelated subcomponents; ‘‘[t]here is no meaningful distinction between
grammar and lexicon. Lexicon, morphology, and syntax form a continuum
of symbolic structures’’ (Langacker 1987 :3). A noun is a symbolic structure for
what is conceptually a thing, i.e., a concept conceived statically and holisti-
cally; a verb is a symbolic structure for a concept that is construed as relational
and mentally viewed across time; an adjective (or other modifier) is a symbolic
structure for a concept construed relationally, but scanned holistically. The
relationship between the construals and their symbolic representations is
established conventionally in particular languages. Likewise, grammatical
morphemes are symbolic structures (Langacker 1991 :3), albeit often more
abstract and ‘‘skeletal’’ semantically, and usually non-referential. Using an
approach compatible with Langacker’s, but different in allowing for universal
semantic parameters, Croft ( 1991 , 2001 ) suggests that parts of speech should
be regarded not as categories of particular languages, but rather as typological
prototypes on which speakers draw in constructing a grammar.
Some linguists prefer to speak not of lexical (major) classes and gramma-
tical (minor, functional) classes, but of open and closed classes. According to
Talmy ( 2000 , I:22), open classes are those that are ‘‘quite large and readily
augmentable relative to other classes’’ and closed classes are those that are
‘‘relatively small and fixed in membership.’’ For him open classes are Ns, Vs,
and in those languages that have them, Adjs. All other linguistic forms are
closed class; these include determiners, prepositions, adverbs, particles, and
some intonation patterns, such as question intonation. Such a division natu-
rally raises the question of how large an open class is, and how to think about
more fine-grained distinctions such as count vs. mass nouns (e.g., knife/two
knives vs. warmth/*two warmths), or transitive vs. intransitive verbs (e.g., hit
vs. come). Talmy’s solution is to treat such categories, and also ‘‘subject,’’
‘‘indirect object,’’ etc., as covert closed class subtypes; these are ‘‘design
features’’ of language and cannot easily be added to. One difficulty with
treating Adjs as open class members is that, according to Dixon ( 1982 ),
in some languages they actually belong to a closed class insofar as
14
For various proposals, see, e.g., Hopper and Thompson ( 1985 ), Langacker
( 1987 ), Croft ( 1991 ), or Pustet ( 2003 ).
14 Theoretical contexts
This idealization was challenged early on in Ross ( 1972 ) who argued that
N, Adj, V should not be regarded as discrete categories; instead, they should
be thought of like the cardinal vowels, what we would now call ‘‘prototypes’’
or idealized clusters of behavioral properties (see, e.g., Rosch 1978 ; Taylor
1997 [1989]), categories that are distinguished by degree rather than kind
(Ross 1972 :326). Individual items will show more or fewer of the criterial
behavioral characteristics. Like Ross, most functional theorists have been
concerned with the gradience not only within but between categories.
16
Some Ns are ‘‘more nouny’’ than others; e.g., house has more nominal
properties than home (e.g., go to the house, *go house, go home [go to the
home has a special meaning, like to the nursing home]). Some members of
Aux are more auxiliary-like than others, e.g., must has more Aux properties
than ought to (You mustn’t smoke, ?You oughtn’t to smoke).
17
Sometimes
degree of categoriality depends synchronically on variety; speakers of
British English treat have to more like an Aux than US speakers (Brit.
Have you to leave?, US Do you have to leave?). Denison ( 2001 ) discusses
several examples involving fun, key, and designer showing that they exhibit
gradience. For example, in That was great fun, fun functions as N. Since Ns
can modify other Ns (as does stone in The stone wall), fun in Those were fun
times can be regarded as an N. But since this syntactic slot between
Determiner (Det) and N is also available for Adjs, there is potential ambi-
guity between fun functioning as N or as Adj in this position. Fun here is
gradient between N and Adj, according to Denison. However, it clearly
functions as Adj in Now let’s think of someone fun, and That was very fun,
The funnest evening. Gradience can be considered a factor motivating
change, and also as the outcome of changes in usage (see above, 1.1.2).
1.2.4 A continuum of productivity
The notion of a continuum is also related to another phenomenon distin-
guishing lexical and grammatical categories, namely, the feature of produc-
tivity. Broadly construed, PRODUCTIVITY is a design feature of language that
allows speakers to produce novel combinations, or it is the speaker’s statis-
tical readiness to produce them (see, e.g., Aronoff 1976 ; Kastovsky 1982 ; Plag
1999 ; Jackendoff 2002 ). Productivity is ‘‘fundamentally concerned with the
ability of a speaker to produce new forms’’ (Bauer 1994 :3357) and is equated
with the frequency, or relative frequency, of a form. In a language like
English, the most productive (or ‘‘default’’) items are grammatical (plural
16
For a critique arguing that gradience within categories is more significant than
between categories, see Aarts ( 2004 ), and, for historical research questions arising
out of this claim, Section 6.2.2.
17
These properties are often called the NICE properties (behavior in negation,
inversion, ‘‘code,’’ and emphatic affirmation) (see, e.g., Palmer 1988 :14ff.).
16 Theoretical contexts
inflection for the noun, third singular present tense and past tense inflections
for the verb), whereas the least productive (‘‘idiosyncratic’’) items are certain
lexical formatives (such as the derivational prefix be- in befriend) and most
lexical items (roots or stems). Although word formation processes such as
clipping, conversion, or blending allow for the production of new lexical items
from existing lexical items (see Chapter 2 ), these processes are relatively
unproductive compared to ‘‘rule-governed’’ grammatical processes (Bauer
1994 :3356). It is important to recognize that productivity is a ‘‘gradient
concept’’ (Bussmann 1996 :s.v. ‘‘productivity’’), ranging from relatively idio-
syncratic patterns (e.g., the voice alternation of the fricative in the pair
north–northern) to relatively regular ones (e.g., the derivation of an adjective
from a noun by -y as in earth–earthy) to highly regular ones (e.g., the deriva-
tion of an agent noun from a verb by -er as in sing–singer). Furthermore, it is
highly dependent on the age of the speaker, and the discourse context
(a standard example of a nonproductive derivational form is nominal -th as
in warmth, depth; however, -th is being used innovatively in on-line discourse,
cf. coolth, greenth, gloomth [Baayen 2003 ]).
18
Typically, a continuum is estab-
lished from ‘‘non- or unproductive’’, to ‘‘semi-productive,’’ to ‘‘productive’’;
this continuum corresponds roughly to the continuum from lexical to gram-
matical. Bauer suggests that ‘‘[t]he converse of productivity... is ‘lexicaliza-
tion’’’ ( 1994 :3355). However, as will be discussed in subsequent chapters,
some derivational morphology is almost as productive as inflectional
morphology, e.g., manner adverbial -ly, as in fortunately (also the nominal
prefix pre- as in pre-season) (Jackendoff 2002 :155).
Of particular interest for studies of language use is the interaction of
productivity with frequency (see, e.g., Baayen and Renouf 1996 ). In fre-
quency studies an issue of central importance is the distinction between type
and token frequency (see, e.g., Bybee 1985 , 2003 ). TYPE FREQUENCY concerns
the number of categories or constructions with which an item cooccurs, e.g.,
how many verbs occur in ‘‘middle alternation’’ constructions such as The
butcher cuts the meat easily The meat cuts easily, or how many verb types
(process, stative, etc.) be going to cooccurs with. By contrast, TOKEN
FREQUENCY concerns the number of instances of a form, e.g., how often
google (verb) or the is used.
19
Type frequency of word formation may
correlate with productivity. However, the correlation is often weak owing
to residues of earlier increases or decreases in type frequency; e.g., -ment has
a high type frequency in contemporary English (it occurs with a very large
number of lexical bases, e.g., government, derailment), but it is currently not
18
Thanks to Anette Rosenbach for this reference.
19
Kastovsky ( 1986 ) uses a slightly different distinction. He says ‘‘productivity’’
encompasses on the one hand the number and type of constraints on the applica-
tion of a word formation rule (he refers to this as ‘‘rule scope’’), and on the other
the frequency of the application of the rule in performance.
Concepts of the lexicon 17
grammar, such as the proposal that the lexeme kill is the representation of
a more abstract structure such as CAUSE BECOME NOT ALIVE
(McCawley 1968 ). Much current research is on predicates and cross-linguistic
evidence for the possible conflation of complex conceptual structures into a
single lexical form (e.g., Talmy 1985 , 2000 ; Jackendoff 1990 , 2002 ), and on
the ways in which a small set of event types (e.g., action, motion, change of
state) link up to a large set of ‘‘roots’’ (or particular lexemes) with idiosyn-
cratic meaning (e.g., Rappaport Hovav and Levin 1998a, 1998b).
One line of research that has been particularly influential is Talmy’s
hypothesis that languages fall into two general typological groups with
respect to the lexicalization of certain motion event types (Talmy 1985 , 2000 ):
(a) Languages like English (and most European languages other than the
Romance languages), Chinese, Finno-Ugric, and Ojibwa that character-
istically code (or ‘‘lexicalize’’) a motion event together with Manner in a
single verb root (or ‘‘lexeme’’), and express the Path of motion in a
‘‘satellite’’ to the verb such as a particle, or affix. In English, The bottle
floated into the cave is the default expression: manner (floated) is con-
flated into the abstract verb MOVE, while the Path (into the cave) is
profiled separately. The bottle entered the cave floating is possible, but is
less colloquial, and is perspectivized differently (the bottle appears to be
treated as agentive). Similar examples in the domain of stative (i.e., non-
motion) location are provided by The rope hung across the canyon (‘the
rope was at the canyon in a hanging manner’), Paint streaked the rug
(‘paint was all over the rug in a streaked manner’).
(b) Languages like Romance, Semitic, Japanese, and Polynesian that char-
acteristically lexicalize a motion event together with Path in a single
lexeme, and express Manner in an associated form. In Spanish, La
botella entro´ flotando en la cueva ‘The bottle entered floating to the
cave’ is the default: motion and Path are conflated (entro´) while manner
(flotando) is profiled.
The first group of languages is said to be ‘‘satellite-framed,’’ the second
kind ‘‘verb-framed.’’ These differences have been shown to have a large
number of consequences for style in narrative and other kinds of discourse.
For example, it is hypothesized that speakers of satellite-framed languages
pay more attention to dynamic activities along a path in narrative, whereas
speakers of verb-framed languages pay more attention to static descriptions
(Berman and Slobin 1994 ). Another hypothesis is that satellite-framed lan-
guages will have a larger inventory of manner of motion verbs (e.g., creep,
glide, clamber, spring) than verb-satellite languages, at least in informal
registers where borrowings are least likely to be used (enter, exit, descend,
borrowed into English from Romance, conflate motion and Path).
The groups are clearly not discrete, not only because of the many borrow-
ings that may occur, but also because of the sheer richness of the ways
Lexicalization 19
in which words can be formed in any language. Furthermore, typological
changes may occur: Latin, for example, had many characteristics of satellite-
framing but its descendants, the Romance languages, are predominantly
verb-framing; however, modern Italian is moving toward satellite-framing
(Slobin 2004 ). Nevertheless, the broad sweep of the claims has fairly wide
acceptance, and the semantic distinctions identified have helped advance
lexical semantics from a marginal to a central role in linguistic theory.
Important synchronic and cross-linguistic claims about lexicalization have
also been made in the narrower domains of quantifiers like all, and modals like
must. It was noticed since Jespersen ( 1917 ) that certain negative meanings have
exclusively phrasal expression cross-linguistically and are not expressible by a
single word-sized unit. In particular, Horn ( 2001 [1989]) and Levinson ( 2000 )
have argued that while cross-linguistically the meanings associated with all,
some, and no(ne) (¼ ‘all not’) are lexicalized in the sense that they have distinct
single word forms, *nall (¼ ‘not all,’ which implies ‘some’) is not. Likewise,
forms expressing necessary, possible, impossible (¼ ‘not possible that’) appear,
but not *nossible (¼ ‘possible that not,’ which implies ‘possible’). The general-
ization is that strong negative scalar items (all, necessary) incorporate negation
morphologically, whereas weak ones (some, possible) do not. Horn and
Levinson argue that because some and possible implicate not all and not
necessary respectively, there is no need for a separate word meaning nall or
nossible. Building on Horn ( 2001 [1989]), van der Auwera ( 2001 ) has shown
that similar, but more flexible, constraints exist for modal verbs. Word order
specialization and even single-word lexicalization of NOT NECESSARY do
occur; e.g., we have must, may, mustn’t (¼ ‘necessary that not’), and also
needn’t (¼ ‘not necessary that,’ which implies may). However, historical drifts
from ‘not necessary that’ or ‘not possible that’ to ‘necessary that not’ and
‘possible that not,’ i.e., to the stronger negative modal, are not infrequent;
hence, modal systems may resemble negative ones with respect to asymmetry
of lexicalization possibilities. Note that in these discussions, ‘‘lexicalized’’
expressions crucially do not include complex fixed multi-word expressions,
but strictly single word-like entities.
1.3.2 Historical perspectives on lexicalization
Historical perspectives on lexicalization will be discussed in more detail
in Chapter 2. For overviews see Brinton ( 2002 ), Lindstro¨ m ( 2004 ),
Himmelmann ( 2004 ), Traugott ( 2005 ). By way of introduction, an intuitive
and broad sense of the term ‘‘lexicalization’’ is that it refers to adoption into
the lexicon, e.g.:
‘‘[T]he adoption of a word into the lexicon of a language as a usual
formation that is stored in the lexicon and can be recalled from there for
use’’ (Bussmann 1996 :s.v. ‘‘lexicalization’’).
20 Theoretical contexts
