New reflections on grammaticalization and lexicalization

Abstract

The essential difference between grammar and lexicon is the following: The grammar is concerned with those signs which are formed regularly and which are handled analytically, while the lexicon is concerned with those signs which are formed irregularly and which are handled holistically. A sign is lexicalized if it is withdrawn from analytic access and inventorized. On the other hand, for a sign to be grammaticalized means for it to acquire functions in the analytic formation of more comprehensive signs. Both processes regularly, but not necessarily involve a reductive component. Consequently, grammaticalization is not the mirror image of lexicalization.

The genesis of members of minor word classes, in particular adpositions and conjunctions, has often been treated as an instance of grammaticalization. However, minor word classes are not necessarily classes of grammatical formatives. In particular, there are more lexical and more grammatical adpositions. For instance, before auf Grund (von) ‘on the basis of’ can ever get grammaticalized to a grammatical preposition, it must first be lexicalized to the lexical preposition aufgrund (von). In this sense, grammaticalization presupposes lexicalization.

Thus, lexicalization and grammaticalization are processes that have much in common and are, to a certain extent, parallel. The mirror image of grammaticalization is degrammaticalization, and the mirror image of lexicalization is folk etymology.

Full text

CLIPP
Christiani Lehmanni inedita, publicanda, publicata
titulus
New reflections on grammaticalization and
lexicalization
huius textus situs retis mundialis
http://www.uni-erfurt.de/
sprachwissenschaft/personal/lehmann/CL_Publ/
New_reflections.pdf
dies manuscripti postremum modificati
28.04.2005
occasio orationis habitae
Symposium "New Reflections on Grammaticalization"; 17.-
20.6.1999, Potsdam
volumen publicationem continens
Wischer, Ilse & Diewald, Gabriele (eds.), New reflections on
grammaticalization. Amsterdam & Philadelphia: J.
Benjamins (TSL, 49).
annus publicationis
2002
paginae
1-18

1
This paper was first presented at the Philosophical Faculty of the University of Tübingen in December
1998. I thank the participants of that guest lecture and the audience of the Potsdam symposium, in particular
Carmen Pensado and Martin Haspelmath, for helpful criticism and suggestions.
New reflections
on grammaticalization and lexicalization
Christian Lehmann
Abstract
The essential difference between grammar and lexicon is the following: The
grammar is concerned with those signs which are formed regularly and which are
handled analytically, while the lexicon is concerned with those signs which are
formed irregularly and which are handled holistically. A sign is lexicalized if it is
withdrawn from analytic access and inventorized. On the other hand, for a sign to be
grammaticalized means for it to acquire functions in the analytic formation of more
comprehensive signs. Both processes regularly, but not necessarily involve a
reductive component. Consequently, grammaticalization is not the mirror image of
lexicalization.
The genesis of members of minor word classes, in particular adpositions and
conjunctions, has often been treated as an instance of grammaticalization. However,
minor word classes are not necessarily classes of grammatical formatives. In
particular, there are more lexical and more grammatical adpositions. For instance,
before auf Grund (von) ‘on the basis of’ can ever get grammaticalized to a
grammatical preposition, it must first be lexicalized to the lexical preposition
aufgrund (von). In this sense, grammaticalization presupposes lexicalization.
Thus, lexicalization and grammaticalization are processes that have much in common
and are, to a certain extent, parallel. The mirror image of grammaticalization is
degrammaticalization, and the mirror image of lexicalization is folk etymology.
1

2
Christian Lehmann
1. Theoretical bases
The purpose of this contribution is to clarify the concepts of 'grammaticalization' and
'lexicalization' in their mutual relationship (cf. Moreno Cabrera 1998). Such an explication
cannot possibly justify all previous uses of these concepts, in particular not all of those
reported or endorsed in Lehmann 1989. Also, to the extent that it tries to be consistent, the
explication necessarily leads to unwonted results.
1.1. Analytic and holistic approaches
Given an object of cognition of some complexity, the human mind has two ways of accessing
it. The analytic approach consists in considering each part of the object and the contribution
that it makes to the assemblage by its nature and function, and thus to arrive at a mental
representation of the whole by applying rules of composition to its parts. The holistic
approach is to directly grasp the whole without consideration of the parts. This can be done
if the object itself is already familiar or if, by its contours or its contextual setting and
function, it bears an essential analogy to some familar object.
The two approaches complement each other in various ways.
1. If confronted with a familiar object, we tend to take the holistic approach; if confronted
with an unfamiliar object, we take the analytic approach.
2. For a given specific object, we can often switch between the two approaches by making
a fresh analysis of what used to be familiar or by disregarding compositional parts in
favour of the function of the whole.
3. A given complex object may only be analyzed in certain parts or aspects, while the
internal structure of other parts remains out of consideration.
To illustrate:
E1. a. X chooses the correct approach to Y.
b. X takes the correct approach to Y.
In E1.a the combination of the relational noun approach with its prepositional dependent,
and the combination of the transitive verb choose with its direct object, are interpreted by
general rules of semanto-syntax.
In E1.b the combination X [takes (Z) approach] to Y constitutes a proper part of the sentence.
Its contour and function are analogous to the simpler construction X approaches Y (in a Z
way).
E1 thus illustrates the above generalizations:

3
New reflections on grammaticalization and lexicalization
approach
com-
plexity level
idiosyncratic regular
holistic
ø
analytic
higher
lexicon
phraseo logy syntax
grammar
ù
morphemicon
mor
word for
phology
mation inflection
lower
S1. Lexicon and grammar
1. The relatively unfamiliar collocation choose .... approach is construed analytically, while
the familiar collocation take ... approach is construed holistically.
2. The collocation choose ... approach could instead be accessed holistically, whereby the
specific contribution of choose would essentially be foregone, and the whole would be
largely synonymous with take ... approach; and again, the collocation take ... approach
could instead be accessed analytically, whereby take would regain a more literal sense
(contrasting, e.g., with abandon), and the resulting constructional meaning would be
slightly different.
3. The holistic approach treats take ... approach as a proper part of the construction, which
it is not in the analytic approach. However, this does not mean that the construction of
E1.b is an unanalyzed whole, since we can still integrate the contributions of each of the
elements in the slots X, Y and Z with the help of general compositional rules.
1.2. Lexicon and grammar
The system of linguistic signs is subdivided into lexicon and grammar. The relationship
between the two components and their organisation in terms of subcomponents is
represented in S1.
On the horizontal axis of S1, the lexicon differs from the grammar. The vertical axis is
associated with the hierarchy of levels of grammatical structure. The latter is, of course, only
partially represented in the lexicon. The most idiosyncratic part of the lexicon is the
morphemicon, which contains all the lexical and grammatical morphemes of the language.
Accessing a collocation XY holistically means treating it as an entry of the inventory, as a
lexical item. If this mode of access to XY gets more prominent in language activity, it is the
initial step of the lexicalization of this sequence.

4
Christian Lehmann
2
All the data and most of the analyses of this language are taken from Schulze-Berndt 2000.
3
CONT continuous, DU dual, PRS present, PST past, SG singular.
Accessing a collocation XY analytically means treating it as a grammatical construction in
which the structural properties of either X or Y or both matter and make a regular
contribution to the pattern. If this mode of access gets more prominent in language activity,
it is the initial step of the grammaticalization of XY.
In the following two sections, we will see that lexicalization and grammaticalization apply
alternatively to a construction, while they apply successively to an item.
2. Lexicalization and grammaticalization as alternatives
2.1. Verb and coverb in Jaminjung
The initial step in the processes of grammaticalization and lexicalization does not yet involve
any noticeable changes in the collocation. So far, those are but alternative modes of treating
the collocation XY. However, they lay the ground for the further fate of XY. To see this, let
us take an example from Jaminjung, a Non-Pama-Nyungan language of Northern Australia.
2
The language has two word classes which are at stake here. One is the class of verbs which
is closed and comprises about 30 members. Understandably, these verbs have a very general
meaning, are highly polysemous and in this resemble the function verbs or even auxiliaries
of more familiar languages. The other class is called coverbs. It is an open class which
comprises such concepts as are covered by verbs and adverbs in more familiar languages.
The coverbs have valence just like the verbs, but they do not take a subject and instead
combine with a verb much like an adverb does. E2 is an example, combining the verb -angga
with the coverb warlnginy.
E2.jirramabuny-anggawarlnginy
two 3.DU-GO.PRS on.foot
3
‘two are walking’ (Schultze-Berndt 2000, DB, D14105)
Now the collocation of verb plus coverb can be accessed either analytically or holistically.
In the former case the verb functions like a grammatical verb. E3 is an example.
E3. jiwayurru buru-mayan ga-gba=biya
bower.bird return-CONT 3.SG-BE.PST=NOW
‘the bower bird was going back and forth then’ (Schultze-Berndt 2000, 2-79)
The meaning of the sentence is construed in a bottom-up fashion by the following
compositional operations. First, the coverb is combined with its (nominal) dependents – none

5
New reflections on grammaticalization and lexicalization
in E3. Then the verb is first combined with its nominal dependents – here, the subject – and
next with the coverb phrase. If they have nominal dependents in common, these and their
roles are unified. In this way, the meaning of the whole is a regular function of the meaning
of the parts and their relations.
In this approach, the collocation of verb and coverb works as a pattern, with two slots to be
occupied by members of two clearly distinct categories, one of which – the verb – constitutes
a structured paradigm. The two slots can be filled essentially in mutual independence. To the
extent that the bulk of the concrete meaning of the whole is contributed by the coverb, the
verb only functions as an aspectual operator which converts the coverb into a finite clause.
In this construction, it is a grammatical verb.
If, instead, the collocation of verb plus coverb is accessed holistically, the verb retains a
concrete meaning, as in E4.
E4. jirrib ga-rdba-ny
married 3.SG-FALL-PST
‘he/she got married’ (Schultze-Berndt 2000, JAM 013)
Here, the verb and the coverb do not each take their dependents. Instead, the complex formed
by the two essentially functions like a derived verb in whose meaning the meaning of the
parts in isolation cannot necessarily be re-identified and which takes dependents as a whole.
Although the collocation bears an outer resemblance to the pattern observed in the former
case, no filling of the slots is possible which could bear a semantically regular paradigmatic
relationship to the one of E3. In this approach, the collocation of verb and coverb functions
as a simple verb, enriching, as it were, the inventory of the verbs. The complex is, thus,
lexicalized.
2.2. The German preposition zu
Similar examples could, of course, be adduced from serial verb constructions all over the
world. Often it is the same verb which gets both grammaticalized to a function verb and,
finally, to an aspectual operator, and in other collocations gets lexicalized by merging with
a contextual component. This will be illustrated with the German preposition zu.
Synchronically, this preposition has a number of uses which vary in the extent to which they
form regular patterns. Originally, this was a local preposition with allative and locative
functions, similar to French à. These two functions appear in E5.
E5. a. Der Prinz begab sich zur Königin. ‘The prince betook himself to the queen.’
b. Der Prinz residierte zu Potsdam. ‘The prince resided at Potsdam.’

6
Christian Lehmann
4
Cf. Diewald (in this vol.) for a detailed analysis of the role of the context in grammaticalization.
The allative use of zu evolves into a purposive one, which in the end gives us the
subordinator of the infinitive. The development can be envisaged as proceeding along a
gradience whose steps are illustrated by E6.
E6. a. Der Prinz begab sich zur Königin. ‘The prince betook himself to the queen.’
b. Der Prinz begab sich zur Jagd. ‘The prince betook himself to the hunt.’
c. Der Prinz begab sich zum Jagen. ‘The prince betook himself to hunting.’
d. Der Prinz entschied sich zum Jagen.‘The prince decided in favor of hunting.’
e. Der Prinz entschied sich zu jagen. ‘The prince decided to hunt.’
At each stage of this evolution, the preposition is sensitive to the syntactic categories that
constitute its context.
4
The dependent is a concrete and an abstract NP, respectively, in E6.a
and b. The noun in the abstract NP is an infinitive in E6. c and d. Finally, the complement
of the preposition is a bare infinitive in e. At the same time, the superordinate verb is one of
locomotion as long as the preposition has an allative or purposive sense. Once the latter does
no more than subordinate an infinitive, the superordinate verb can be a complement-taking
verb. In the end, such verbs may even require zu as a marker introducing the dependent
infinitive. At this endpoint of the grammaticalization process, zu is but an obligatory slot
filler in a construction which is formed by compositional rules of syntax.
The uses of zu which form the chain leading from E5.a = E6.a to E6.e instantiate productive
patterns. This is not so with the locative use appearing in E5.b. The combination of locative
zu with town names is obsolete. If a toponym which is not a town name, such as Capri,
Hessen, Dänemark, is substituted for Potsdam, the sentence becomes outright
ungrammatical. The combination of locative zu with common nouns is illustrated in E7.
E7. a. Der Prinz war zu Hause. ‘The prince was at home.’
b. Der Prinz kam zu Pferde. ‘The prince came on horseback.’
c. Der Prinz siegte zu Wasser und zu Lande.
‘The prince triumphed on land and sea.’
Each of the collocations of zu with its dependent in E7 is lexicalized. In E7.a, we cannot
substitute Hause by Hotel; Esel ‘donkey’ instead of Pferde in E7.b, and Fluß/Ufer
‘river/bank’ instead of Wasser/Lande in E7.c are impossible. The collocations do instantiate
a pattern, viz. the one illustrated by E5.b; but the pattern is obsolete. Consequently, the
phraseologisms of E7 are but remnants of an earlier pattern. In regular locative prepositional
phrases, zu is replaced by other prepositions.
2.3. Grammaticalization of a construction

7
New reflections on grammaticalization and lexicalization
5
As Meillet (1915:170) says: “c'est le rôle dans la phrase qui décide de tout”. Cf. also Bybee et al. 1994:
11.
The examples show that one cannot properly say that a given element as such is either
grammaticalized or lexicalized. Instead, it is the construction of which the element is a
constituent which may embark on either course.
5
If this is so, then the grammaticalization of
a construction does not entail the grammaticalization of any of its component elements.
Consider, for example, the construction consisting of a verbum dicendi as superordinate verb
and a subordinate clause, as in E8.b.
E8. a. Irvin apologized, he didn't hit me on purpose.
b. Irvin said he didn’t hit me on purpose.
It seems appropriate to say – as it has been said for at least a century – that the collocation
of a verbum dicendi and a sentence specifying the content of the communication, as it
appears in E8.a, has been grammaticalized into a complex sentence in E8.b. No formative
is visible in E8.b which would have specifically undergone grammaticalization in this
process.
E9 translates E8 into German.
E9. a. Erwin entschuldigte sich; er habe mich nicht absichtlich getroffen.
b. Erwin sagte, er habe mich nicht absichtlich getroffen.
Here, the sentence rendering the content of the speech act is in the subjunctive present in
both cases. In E9.a, the subjunctive expresses that the speaker does not vouch for what he
is saying. In E9.b, it is triggered by the governing verb. Consequently, the subjunctive
becomes more grammaticalized in this development.
The analogy between the English and the German case warrants the generalization that the
grammaticalization of a particular formative is but a by-product of the grammaticalization
of a construction. If there is an element that mediates the relation between the constituents
of a construction, then grammaticalization of the construction will involve
grammaticalization of this element. But if there is no such element present,
grammaticalization may proceed, anyway.
Those who are familiar with my earlier work on grammaticalization will notice that this
implies a slight extension of the concept. The traditional conception, which centers around
the grammaticalization of a linguistic sign, sees this in the intersection of a set of
paradigmatic and syntagmatic relations. This, of course, entails the existence of a
construction that the sign in question is a part of and that is grammaticalized together with
the latter (cf. Lehmann 1995[S], §2 and 1995[T]:175-178). What I propose here is to apply

8
Christian Lehmann
6
Kortmann & König 1992:684 arrange some deverbal English prepositions on a continuum from least to
most grammatical.
lexical grammatical
category example meaning category example meaning
noun posesión possession pronoun suyo his
adjective rojo red pro-adjective tal such
verb existir exist auxiliary estar be
adverb atrás behind deictic adverb ahí there
preposition tras behind gramm. prep. de of
conjunctio
n
mientras while subordinator que that
T1. Lexical and grammatical members of word classes in Spanish
the criteria of paradigmatic and syntagmatic autonomy to a construction regardless of
whether it contains a constituent in which the symptoms of grammaticalization crystallize.
This is a small and controlled extension which avoids the undesirable consequence that
anything which enriches the grammar must be called grammaticalization. The extension is
of relevance for the analysis of isolating structures, on whose grammaticalization much more
empirical work is necessary.
3. Lexicalization and grammaticalization in succession
3.1. Lexical and grammatical members of word classes
Up to now, we have viewed grammaticalization and lexicalization as two alternatives which
may apply to a given construction. However, there is yet another sense in which the two
processes complement each other. Before we can turn to it, the theory of word classes needs
some clarification.
Word classes are sometimes divided into lexical and grammatical word classes. It is, for
instance, assumed that nouns, adjectives and verbs form lexical word classes, while
prepositions and conjunctions form grammatical word classes. As a matter of fact, there are
lexical and grammatical words in each of the word classes, as illustrated in T1 from Spanish
(the English translations illustrate the same point).
6

9
New reflections on grammaticalization and lexicalization
7
The category of the particle is especially often associated with the notion of grammatical word or
morpheme.
8
This insight is already foreshadowed in the Cours de linguistique générale. Saussure (1916[1985]:186)
writes: ‘on attribue généralement les prépositions à la grammaire; pourtant la locution prépositionnelle en
considération de est essentiellement lexicologique, puisque le mot considération y figure avec son sens
propre.’
9
‘Decategorialization’ is presented in Hopper & Traugott 1993, ch. 5.3 as something essential in gram-
maticalization. However, this should not be interpreted as ‘shift from major to minor category’, but rather as
‘shift from content words to function words’ (Haspelmath 1998:329).
10
Cf. Haspelmath 1998 for critical discussion.
The subdivision in T1 shows that word classes do not differ in that some are lexical and
others are grammatical. Instead, the criterion of lexical vs. grammatical is independent of the
word classes and yields two subclasses of each of them. Needless to say, no sharp boundary
between these two subclasses is intended.
By consequence, it is not the case that the so-called minor parts of speech have something
particularly grammatical about them.
7
Therefore the transition of, e.g., a relational noun into
a preposition or a conjunction does not amount to the grammaticalization of the former, as
is so often assumed. If, for instance, a relational noun such as Span. base appears in a
preposition like a base (de) ‘on the basis (of)’, this is often called grammaticalization of the
noun base to the preposition a base de. In reality, however, the appurtenance of any
linguistic unit to a word class – preposition in the case at hand – implies first and foremost
its appurtenance to the inventory, i.e. to the lexicon.
8
The genesis of a preposition like a base
(de) is therefore, first of all, a process of lexicalization of this sequence. Once such new
lexical items have been created, they can undergo grammaticalization.
If grammaticalization changes a lexical element of a given category into a grammatical one,
then it follows that grammaticalization by itself does not touch the syntactic category or may,
at any rate, leave it untouched.
9
These are cases of grammaticalization which cannot be
construed as reanalysis.
10
3.2. Lexicalization and grammaticalization in Spanish
In what follows, we consider the relationship between lexicalization and grammaticalization
in three different areas of Spanish, viz. prepositions, conjunctions and verbs with
prepositional government.
3.2.1. Complex prepositions

10
Christian Lehmann
11
Cf. Raible 1992, ch. I.1 and, in particular, p. 11f on the strength of the subtype a N de.
12
This is, of course, not to say that all Spanish prepositions originate in this way. There are also
deadverbial prepositions (on which s. Meyer-Lübke 1899:159-164) such as dentro ‘inside’ (< Vulgar Latin
de intro), which initially govern a direct complement (dentro la casa ‘inside the house’), but from the second
half of the the 13
th
century on govern their complement by means of de (dentro de la casa).
form meaning
Old Castilian Mod. Castilian
des de desde since
baxo de bajo below
a cabo de cabe beside
ante de ante before
face a hacia towards
T2. Complex prepositions in Spanish
Just like other languages, Spanish possesses a number of structural types of complex
prepositions.
11
Of these, only the type illustrated in E10 is of present interest.
E10. Tenemos que venderlo por debajo del precio. ‘We have to sell it under price.’
Prepositions of this structure have been formed since the Old Castilian stage. T2 shows a
selection of Spanish prepositions which were complex at the stage of Old Castilian and are
mere secondary prepositions nowadays.
12
All the prepositions of the first column of T2 have it in common that their last element is a
primary preposition, namely either de or a (cf. Meyer-Lübke 1899:295-298). Only when the
complex changes into a (secondary) simple preposition, the primary preposition may
disappear.
We see that the syntactic pattern of the formation of complex prepositions is based on the
combination of two components, as schematized in S2.a. The first is a semantically specific
expression, to be called semantically specific relator (SSR) in what follows. The second is
a PrepP which, in turn, is formed with a primary, thus, semantically empty, preposition. The
latter only serves the structural function of forming a PrepP. Once this is guaranteed, the
syntactic nature of the SSR and its syntagmatic relation to the PrepP are of secondary
importance. In the SSR slot we find prepositions (de, ante), adverbs (bajo) and relational

11
New reflections on grammaticalization and lexicalization
13
Cf. Vincent 1997. He derives Ital. dopo cena ‘after dinner’, by the reanalysis formalized in S2 (with Prep
instead of SSR in the first position), from Vulgar Latin [de [ post cenam ] ].– It should be obvious in general
that the univerbation of bc in abc presupposes a bracketing a[bc], and consequently, in particular, that
lexicalizations of the kind observable in T2 presuppose the kind of rebracketing shown in S2. It is not clear
how Haspelmath (1998:330-333) avoids this conclusion.
14
Intervocalic d may get lost more generally from the 15
th
century on; thus puede ‘can’ may appear as pue.
15
This development would go in the opposite direction of the one mentioned in fn. 12. In the case at hand,
an alternative analysis is possible, as pointed out by M. Haspelmath (p.c.). Namely, the phrase bajo de NP
never passes through S2. Instead, bajo at first governs a prepositional complement; later, government
becomes direct, so that de is dropped. This analysis, while plausible in the cases at hand, is obviously not
available for the cases of T2.
a. input [ [ SSR [ primary prep [ NP ] ]
PrepP
]
PrepP
b. transition [ [ SSR primary prep ] [ NP ] ]
PrepP
c. output [ secondary prep [ NP ] ]
PrepP
S2. Genesis of complex prepositions by reanalysis
nouns (face); and the syntactic relation between the two syntagms varies accordingly. If the
SSR is a preposition or a relational noun, we have government; in the other cases, we have
modification or apposition. As the example of des de (< Vulgar Latin de ex de) in T2 shows,
this pattern is of old age. Comparison with French dès reveals that it goes back to
Proto-Romance.
It is probably not too important that the internal syntax be entirely correct here, because the
SSR is not meant to function as a compositional constituent of a syntactic construction, but
to create a semantically specific form of prepositional subordination. Since language is a
goal-directed activity, we may assume that the reanalysis shown in S2.c is already targetted
with the formation of those syntactically complex expressions.
The reanalysis goes hand in hand with the lexicalization of the complex preposition, since
step S2.b subtracts the formation from the rules of syntax.
13
In the initial phase, the primary preposition was needed for its structural function. After
lexicalization, this function is integrated into the complex consisting of the SSR plus
preposition. The internal structure of the latter is no longer relevant. It may either be blurred
by phonological attrition, as in cabe and hacia;
14
or the primary preposition may be dropped,
as in bajo.
15

12
Christian Lehmann
16
See Raible 1992 for a cross-linguistic survey and a universal theory of clause linkage.
17
Cf. Meyer-Lübke 1899:611f. Meillet 1915 is among the first to analyze the genesis of Romance
conjunctions in terms of grammaticalization.
form meaning
en la medida en que (c. ind.) to the extent that
a pesar de que (c. ind.) despite the fact that
a no ser que (c. subj.) lest
después de que (c. ind./subj.) after
pese a que (c. ind.) although
con tal que (c. subj.) provided
siempre que (c. subj.) as long as
para que (c. subj.) in order that
aunque (c. ind./subj.) though
porque (c. ind.) because
T3. Secondary conjunctions in Spanish
3.2.2. Complex conjunctions
Just like other languages, Spanish possesses a number of structural types of complex
conjunctions.
16
Of these, only the type illustrated in E11 is of present interest.
E11. No parece mal que los españoles tengan sus patatas, con tal que nosotros tengamos
nuestras papas. ‘It seems o.k. for the Spanish to have their pommes de terre,
provided we can have our potatoes.’
T3 shows a mixed selection – this time, at the synchronic level – of complex and other
secondary conjunctions of Spanish.
All these conjunctions have it in common that their last element is a primary conjunction,
viz. que.
17
Only when the complex changes into a (secondary) simple conjunction, such as
como, ‘as’ and mientras ‘while’, que disappears. For several of these complex conjunctions,
the subordination is syntactically completely regular. For instance, in a pesar de [que S],
después de [que S], pese a [que S], we can substitute a concrete NP such as mis esfuerzos

13
New reflections on grammaticalization and lexicalization
18
Cf. Kortmann 1996 for the various combinations found in European languages and Herlin 1999 for
complex temporal conjunctions in Finnish.
19
The same schema applies, of course, to the formation of conjunctions in other Romance and, mutatis
mutandis, numerous further languages. Cf. Harris & Campbell 1995:288 for Romance.
a. input [ [ SSR [ SR [ S ] ]
NP
]
PrepP
]
AdvP
b. output [ [ sec. conj ] [ S ] ]
AdvP
S3. Genesis of complex conjunctions by reanalysis
‘my efforts’ for the constituent [que S]. This state is represented in S3.a (where SR stands
for ‘[generic] subordinator’).
We see that the syntactic pattern of the formation of complex conjunctions is based on the
combination of a SSR with a subordinate clause which, in turn, is formed with a semantically
empty conjunction. The latter only serves the structural function of subordinating the clause.
Once this is guaranteed, the syntactic nature of the SSR and its syntagmatic relation to the
subordinate clause are of secondary importance. In the SSR slot we find prepositions,
adverbs and combinations thereof; and the syntactic relation between the two syntagms
varies accordingly.
18
If the SSR ends in a preposition, it governs the subordinate clause,
otherwise it bears a modifying or appositive relation to the latter.
19
The conjunction thus developped functionally contains a subordinator and by virtue of this
directly takes a clause as its complement. This state is symbolized in S3.b. In this last phase,
the subordinator que, which remains visible in porque and aunque, may be suppressed, as
in como and mientras.
At the last stage of the evolution of complex prepositions and conjunctions, the structural
element which subordinates the complement and which differs between PrepPs and
subordinate clauses, disappears. As a result of this, there are several particles such as como,
mientras which function both as preposition and as conjunction.
Incidentally, the mood to be observed with complex conjunctions cannot be derived from
their constitution, but is simply the same which is used in the respective subordinate clause
types if they are introduced by a simple conjunction. This is further evidence for the
conception that the passage through S3 is a goal-directed process, where new collocations
are fitted into a given schema.

14
Christian Lehmann
form meaning
acabar/terminar por end with
creer en believe in
parecerse a resemble
asombrarse de wonder at
acabar de finish
abusar de abuse
T4. Prepositional government in Spanish
3.2.3. Prepositions governed by verbs
Verbs govern their complements in different syntactic functions. Among these is the PrepP
as complement. Just as a verb governs the case of its complement, it may govern the specific
preposition of this PrepP. T4 contains some Spanish examples.
Just as in complex prepositions, those prepositions which are used to govern the complement
are exclusively primary prepositions. No Spanish verb governs any of the prepositions of T2.
This is, thus, completely parallel to the complex prepositions. We might therefore feel
tempted to speak of a lexicalization of verb-preposition collocations in T4. However, the
combination here remains discontinuous. Neither is there any evidence for a reanalysis,
analogously to the previous cases, whereby the verb would form a constituent together with
the governing preposition. Therefore the traditional description is to be preferred, whereby
the verb is a lexical unit in itself and determines the occurrence of a particular preposition
in its complement. Anyway, this property of the verb is a lexical-grammatical property, just
as if the preposition were part of its lexeme.
4. Reanalysis, grammaticalization and lexicalization
The reanalyses shown in S2 and S3 destroy a regular syntactic construction with no
compensation at the syntactic level. The reanalysis therefore entails a loss in compo-
sitionality. It is the essential step in the lexicalization of complex prepositions and
conjunctions. It is true that, in the cases reviewed, a new preposition or conjunction evolves
by reanalysis. However, as was said in the beginning, this is not a case of grammaticalization,
because the particle thus developped is not a grammatical element. It could be further
grammaticalized, such as Latin de, ad, in were grammaticalized to Spanish de, a, en. This
process, however, does not involve further reanalysis.

15
New reflections on grammaticalization and lexicalization
20
studied in G. Diewald's contribution to a workshop of the 21. Annual Meeting of the DGfS, 1999
4.1. Grammaticalization vs. lexicalization
Every monomorphemic unit is, by definition, already in the lexicon and therefore cannot be
lexicalized. Only complex units may be lexicalized. Again, relatively few morphemes and
even fewer complex units are contained in the grammar. Morphemes, and also complex
units, may therefore be grammaticalized.
Complex units may be grammaticalized without having been lexicalized. For instance, the
combination of a preposition with its governed case, or the combination of a conjunction
with a mood, may be grammaticalized. These constellations are usually not analyzed as
discontinuous linguistic units (cf., however, Touratier 1979). This would, in fact, presuppose
a reanalysis and ensuing lexicalization of the combination.
Grammaticalization involves an analytic access to a unit, lexicalization involves a holistic
access to a unit, a renunciation of its internal analysis. Both processes do not concern signs
in isolation, but signs in their paradigmatic and syntagmatic relations. However, this is just
where the essential differences between the two processes are.
Let [XY]
Z
be a complex construction which undergoes grammaticalization or lexicalization.
Then the differences between the two processes consist in two aspects. First, in
grammaticalization there may be a constituent of Z, e.g. X, which is the focus of the process
and which is changed into a grammatical formative by it. In lexicalization, there is no such
constituent; the lexicalization affects Z as a whole. From this it follows that lexicalization
necessarily concerns an internally complex unit, whereas we may reasonably speak of
grammaticalization even with respect to simple units.
Second, in grammaticalization the internal relations of Z become more strict and constrained.
This regards, in particular, the relation between X and Y or between X and Z. Again, in
lexicalization the internal relations of Z become irregular and get lost.
A consequence of this explication of the notion of lexicalization is that the coalescence of
two grammatical morphemes must be called lexicalization. Here are a couple of examples.
In Spanish desde, appearing in T2, the grammatical prepositions de, ex and de are combined
to a new preposition. In English himself (with the other inflected forms), the accusative of
the personal pronoun is combined with the semigrammatical morpheme self to yield the
reflexive pronoun. From this perspective, the evolution of German möchte into a new lexeme
(infinitive möchten)
20
is a merger of the semigrammatical lexeme mögen ‘may’ with the
inflectional category of the subjunctive II into a new attenuated volitive meaning.

16
Christian Lehmann
component
complex unit
lexicon grammar
access holistic analytic
structure opaque transparent
function vs. structure irregular compositional
process
²
lexicalization
÷
folk etymology
T5. Lexicalization
4.2. Lexicalization and its converse
What was said on lexicalization may be summarized in T5.
The horizontal axis of T5 is the horizontal axis of S1. Lexicalization is a process constantly
at work in ordinary language activity. T5 shows that the inversion of lexicalization is not
grammaticalization. Bestowing structure onto a hitherto opaque expression is not an
automatic ingredient of language activity, but demands an enhanced measure of creativity.
The operation is called folk etymology (cf. Untermann 1975) and is by magnitudes rarer than
lexicalization.
A final terminological remark is necessary. The adjective lexical has two meanings in
linguistics, 1) belonging to the inventory, 2) having a specific, concrete meaning. In the latter
sense, lexical is opposed to grammatical, as displayed in S1. In the former sense, however,
both words with a concrete meaning and grammatical formatives belong to the inventory. In
particular, the morphemicon (core of the lexicon) contains both the lexical and the
grammatical morphemes of a language. Lexicalization is a process in which something
becomes lexical in the first of the two senses. The term idiomaticization has essentially the
same meaning. Lexicalization as a process in which something becomes lexical in the second
sense would be the same as degrammaticalization, to which we turn in the following section.
4.3. Grammaticalization and its converse
What was said on grammaticalization may be summarized in T6.

17
New reflections on grammaticalization and lexicalization
level higher lower
complex unit sentence word stem
function vs. structure iconic arbitrary
unit concerned word morpheme feature
manipulation free obligatory
meaning specific abstract
process
÷
grammaticalization
²
degrammaticalization
T6. Grammaticalization
The horizontal axis of T6 is the vertical axis of S1. Grammaticalization is a process
constantly at work in ordinary language activity. T6 shows that the inversion of
grammaticalization is not lexicalization. Giving autonomy to a hitherto dependent expression
is not an automatic ingredient of language activity, but demands an enhanced measure of
creativity. The operation is called degrammaticalization (cf. Ramat 1992) and is by
magnitudes rarer than grammaticalization.
5. Summary
Both lexicalization and grammaticalization are reductive processes which constrain the
freedom of the speaker in selecting and combining the constituents of a complex expression.
Insofar, both processes can be regarded as a transition of an expression from parole into
langue. This is in consonance with the conception of langue as the language system whose
semantic subsystem consists of the lexicon and the grammar. Lexicalization and
grammaticalization are the two janus-faces of the creation of the language system in parole,
of the Versprachlichung of the world.
Grammaticalization and lexicalization are not mirror images, but orthogonal to each other.
Both are reduction processes (cf. Lehmann 1989), but in a different sense. Gramma-
ticalization reduces the autonomy of a unit, shifting it to a lower, more strictly regulated
grammatical level, more precisely, into the right lower corner of S1. Lexicalization reduces

18
Christian Lehmann
the inner structure of a unit, shifting it into the inventory, more precisely, into the left lower
corner of S1.
While we may reasonably speak of lexicalization only with respect to complex units,
grammaticalization concerns a complex unit and may simultaneously affect in particular one
of its constituents. The latter then evolves into a (more) grammatical formative. Such a
unitary constituent is created by lexicalization to begin with. Insofar, lexicalization plays a
role as the first phase, or perhaps rather preparatory phase, of grammaticalization. Again, it
is not excluded that lexicalization and grammaticalization occur jointly in a given case.
English wanna and gonna would be cases in point. In the former, the combination of a
lexical and a grammatical morpheme lexicalizes to a modal, in the latter, the combination of
semi-grammaticalized going with a grammatical morpheme is lexicalized and further
grammaticalized.
We have seen that prepositions and conjunctions come about not by grammaticalization, but
by lexicalization. Once they have come into existence, they may then be grammaticalized.
Lexical change, however, is much more ephemerous than grammatical change. From among
all the new prepositions and conjunctions, only a fraction is grammaticalized. All the others
are abandoned and replaced by other neologisms. Those numerous complex prepositions and
conjunctions which constantly come and go do not indicate incomplete grammaticalization
processes, but are simply products of lexical change.
References
Bybee, Joan L. & Perkins, Revere & Pagliuca, William 1994, The evolution of grammar.
Tense, aspect and modality in the languages of the world. Chicago & London:
University of Chicago Press.
Harris, Alice C. & Campbell, Lyle 1995, Historical syntax in cross-linguistic perspective.
Cambridge: Cambridge University Press.
Haspelmath, Martin 1998, "Does grammaticalization need reanalysis?" Studies in Language
22:315-351.
Herlin, Ilona 1999, The lexicalization of Finnish conjunctions. Helsinki: University of
Helsinki (unpublished paper).
Hopper, Paul J. & Traugott, Elizabeth C. 1993, Grammaticalization. Cambridge: Cambridge
University Press (Cambridge Textbooks in Linguistics).
Kortmann, Bernd 1996, Adverbial subordination. A typology of adverbial subordinators
based on European languages. Berlin & New York: W. de Gruyter.
Kortmann, Bernd & König, Ekkehard 1992, "Categorial reanalysis: the case of deverbal
prepositions." Linguistics 30:671-697.

19
New reflections on grammaticalization and lexicalization
Lehmann, Christian 1989, "Grammatikalisierung und Lexikalisierung". Zeitschrift für
Phonetik, Sprachwissenschaft und Kommunikationsforschung 42:11-19.
Lehmann, Christian 1995, "Synsemantika". Jacobs, Joachim et al. (eds.), Syntax. Ein
internationales Handbuch. Berlin: W. de Gruyter (Handbücher der Sprach- und
Kommunikationswissenschaft, 9/II); 1251-1266.
Lehmann, Christian 1995, Thoughts on grammaticalization. Unterschleissheim: L
INCOM
Europa (L
INCOM
Studies in Theoretical Linguistics, 1).
Meillet, Antoine 1915, “Le renouvellement des conjonctions.” Annuaire de l'École Pratique
des Hautes Études, section historique et philologique. Reprint: Meillet, Antoine 1921,
Linguistique historique et linguistique générale. [Tome I].} Paris: Klincksieck; 159-174.
Meyer-Lübke, Wilhelm 1899, Romanische Syntax. Leipzig: O.R. Reisland (Grammatik der
Romanischen Sprachen, 3).
Moreno Cabrera, Juan Carlos 1998, "On the relationships between grammaticalization and
lexicalization." Giacalone Ramat, Anna & Hopper, Paul J. (eds.), The limits of
grammaticalization. Amsterdam & Philadelphia: J. Benjamins (Typological Studies in
Language, 37); 211-227.
Raible, Wolfgang 1992, Junktion. Eine Dimension der Sprache und ihre
Realisierungsformen zwischen Aggregation und Integration. Heidelberg: C. Winter
(Sitzungsberichte der Heidelberger Akademie der Wissenschaften.
Philosophisch-historische Klasse, 1992/2).
Ramat, Paolo 1992, "Thoughts on degrammaticalization." Linguistics 30:549-560.
Saussure, Ferdinand de 1916[1985], Cours de linguistique générale. Publié par Charles
Bally et Albert Séchehaye avec la collaboration de Albert Riedlinger. Édition critique
préparée par Tullio de Mauro. Paris: Payot.
Schultze-Berndt, Eva 2000, Simple and complex verbs in Jaminjung. A study of event
categorization in an Australian language. Nijmegen: Max-Planck-Institute for
Psycholinguistics (MPI Series in Psycholinguistics).
Touratier, Christian 1979, "Accusatif et analyse en morphèmes." Bulletin de la Société de
Linguistique de Paris 74(1):43-92.
Untermann, Jürgen 1975, "Etymologie und Wortgeschichte." Seiler, Hansjakob (ed.),
Linguistic workshop III. Arbeiten des Kölner Universalienprojekts 1974. München: Fink
(Structura, 9); 93-116.
Vincent, Nigel 1997,
?
On the origin of prepositions”. Paper delivered to the Philological
Society, Manchester, 21 November 1997.