Beruflich Dokumente
Kultur Dokumente
BERND HEINE
Institut für Afrikanistik, Universitat zu Köln
ABSTRACT
1. Introduction1
exclude them from language [...]" (Hockett 1955: 17). The fact that the
open-class nature of at least one component of language structure, the lexi-
con, is hard to reconcile with such a position was usually explained away in
some way or other. The following proposal by Halliday is an attempt to
deal with this problem:
Any part of linguistic form which is not concerned with the operation of
closed systems belongs to the level of lexis. The distinction between closed
system patterns and open set patterns in language is in fact a cline; but the
theory has to treat them as two distinct types of pattern requiring different
categories. For this reason General Linguistic theory must here provide
both a theory of grammar and a theory of lexis, and also a means of relat-
ing the two (Halliday 1961: 247).
There is no doubt that approaches in terms of discrete categorization
are economical, practical, as well as descriptively and pedagogically useful
since they provide easy access to language structure. On the other hand, it
should be made clear that such approaches may entail a rigid simplification
or even distortion of the data to be analyzed. Thus, while I do not wish to
challenge the usefulness of such approaches, I will argue that they capture
only one aspect of language structure, and that a more comprehensive
theory of language also has to account for linguistic phenomena which are
hard to reconcile with the characteristics listed above. Some of these
phenomena will be the subject of the present paper; they have to do with:
probabilities rather than with a fixed rule behavior,
continuity rather than with discrete categorization, and with
vagueness and fuzzy boundaries rather than with clear-cut boundaries.
Attention will be drawn to the presence of a specific kind of linguistic
structure which differs drastically from conventional types of categories.
The notion "chain" was introduced by Claudi and Heine (1986) and
adapted by Craig (1991) and Heine, Claudi and Hünnemeyer (1991a: Ch.
8.4) to refer to such structures. Instead of "chain", the term "continuum"
(Heine & Reh 1984: 15; Kölver 1984) or "grammaticalization scale"
(Lehmann 1982; Heine & Claudi 1986) was used in earlier works. 2 It is
hoped that the following discussion will make it clear that "chain" is in fact
a more appropriate label; Colette Craig describes this phenomenon in the
following way:
The term 'chain' is used to evoke the step-by-step nature of the gram-
maticalization process, which is most directly observable in the pairing of
GRAMMATICALIZATION CHAINS 337
two morphemes through a scenario of change, creating links with the inter-
nal structure outlined below:
[SOURCE ... pathway ... OUTCOME]
Chaining happens when the outcome of a link becomes the source compo-
nent of another link [...]. Incorporating the notions of grammaticalization
and chaining into the descriptive grammar of a language has the effect of
adding both a dynamic and a diachronic dimension to the task. In the syn-
chronic study of a particular language, the phenomenon of 'chaining' sur-
faces in morphemes related by polysemy which can be interpreted as traces
of links of various chains (Craig 1991: 455-456).
In the following paragraphs it will be argued that grammaticalization
chains constitute a type of linguistic category that has not been accounted
for adequately in linguistic theory. As has been argued elsewhere (Heine,
Claudi & Hünnemeyer 1991a; 1991b), such categories are mainly the result
of context-induced reinterpretation and the interaction between cognitive3
and pragmatic forces.
also be used in the case of subject participants which are neither human nor
wilful but rather inanimate. The extension of -yyéú from human partici-
pants ('the man wants ...') to inanimate ones ('the tree wants ...') has had
remarkable implications for the meaning of the auxiliary, as (5) shows.
(5) k-é-yyéú l-cáni n-é-uróri
k-3sG- M-tree.NOM NAR-3sG-fall
'the tree almost fell'
Since an inanimate participant is incapable of wilful behaviour, the
auxiliary underwent what has been called context-induced reinterpretation
by Heine, Claudi and Hünnemeyer (1991a): a proposition like 'the tree
wanted to fall' is interpreted as meaning 'the tree almost fell'. With this
interpretation, an aspectual notion was introduced: in examples such as (5),
the auxiliary expresses the sense 'having almost reached the situation
described by the main verb'. This new sense has been conventionalized to
an aspectual category, and as such, it is no longer confined to animate sub-
jects.
The next stage is one where the aspectual meaning also affects the
morphosyntax — with the effect that -yyéú is grammaticalized to an aspect
morpheme which I propose to call the ALMOST-aspect. This development
has the following implications:
(a) The lexeme -yyéú is grammaticalized in its 3rd person singular
form to an invariable aspect marker (k)eyyéú which may no longer be
inflected for person or aspect, as example (6) shows.
(6) (k-)eyyéú a-ók nánu kulέ
1sG-drink I.NOM milk
T was about to drink milk'
(b) The following main verb may no longer take the "narrative"
marker n- but rather appears in the imperfective aspect, for the following
reason: the marker n- may not be prefixed to the first verb of a sentence
and, with the grammaticalization of -yyéú to an aspect marker, the main
verb is now the first verb in the sentence — with the effect that the structure
[auxiliary — main verb] is reanalyzed as a structure [aspect marker —
verb].
(c) In accordance with the verb-initial syntax of Charnus, the subject
has to follow either the auxiliary verb or the main verb. 7 With the loss of its
verbal status, (k)eyyéú may no longer be followed by the subject noun
340 BERND HEINE
phrase, rather the subject has to be preceded by the main verb. Example
(7a) is therefore ungrammatical and has to be replaced by (7b).
(7) a. *(k)eyyéú ninye e-ók kulέ
b. (k)eyyéú e-ók ninye kulέ
3sG-drink s/he.NOM milk.ABS
's/he almost drank milk'
The grammaticalization of (k)eyyéú has, however, gone even one step
further. In the above examples, I have given a simplified translation of the
sentences in the ALMOST-aspect. Thus, a more appropriate translation of
(4) would be 'the old man almost fell (but he didn't)', and of (6) T almost
drank milk (but I didn't)', etc. While the focal sense of (k)eyyéú in exam-
ples such as (4) or (6) in fact is 'having almost reached the situation
described by the main verb', there is also a general inference to the effect
that that situation is actually not reached, that is, the ALMOST-aspect has
a non-focal sense 'the situation described by the main verb is actually not
reached'. 8 Now in certain contexts, this non-focal sense may turn into a
focal one. This is the case in particular in replies to certain polar questions
where situations in the past are implied, as in (8).
(8) i-túm-o m-partút? (k)éyyeu a-túm
2sG-get-PFv F-woman.ABS 1SG-get
'Did you get a wife? No (but I almost did)'
In such contexts, (k)eyyéú9 assumes the function of a negation marker.
Thus, (9) is similar in meaning to (10), where the perfective negation
marker eitú is used.
(9) έ-comɔ l-páyyan n-káŋ? éyyeu
3sG-go.PFv M-elder.NOM F-home?
'Did the old man go home? No (but he almost did)'
(10) e-comɔ l-páyyan n-káŋ? eitú
'Did the old man go home? No (not yet)'
These are but the most salient uses, or stages, as I will refer to them,
in the use of the lexeme -yyéú. The different meanings of this form may be
described as an example of polysemy, although one might argue that the
fact that -yyéú behaves as a lexical category on the one hand, and as a gram-
matical one on the other, makes it a doubtful candidate for polysemy (cf.
Lyons 1977: 561). I will return to this issue in Section 7. The behavior of
these stages is summarized in Table 1, where Stage I represents the least
GRAMMATICALIZATION CHAINS 341
STAGE
Parameter I II III IV V
Example in text (1,2) (3) (4,5) (6,7) (8,9)
1 Conceptual wilful, wilful, inanimate inanimate inanimate
nature of human human or animate or animate or animate
the subject
2 Conceptual inanimate clausal clausal clausal clausal
nature of the or animate proposition proposition proposition proposition
"complement" patient
3 Meaning volition volition aspect aspect negation
4 Morphosyntactic full verb auxiliary auxiliary invariable invariable
status like like particle particle
5 Status of the noun verb in verb in imperfec- imperfec-
"complement" phrase n-tense n-tense tive tive
6 Time spectrum future, future,
of the propos- present present present present
ition or past or past or past or past past
7 Position of before before before after after
the subject "comple- or after or after main main
ment" main verb main verb verb verb
3. Discussion
(e) As we have seen above, there is no meaningful way of dividing the var-
ious uses of -yyéú into, say, a verbal and an aspectual component or
category, neither on the basis of discrete ("classical") categorization
nor on the basis of prototype logic. Rather the entire range of uses con-
stitutes a single linguistic entity, one that embraces both lexical and
grammatical uses. Grammaticalization chains thus appear to form lin-
guistic categories sui generis which are elusive of orthodox approaches
to categorization.
4. Approaches to categorization
family resemblance and proposed two measures for each category member-
ship: its prototypicality rating and its rating of the number of common attri-
butes, and they came to the conclusion that the two measures are closely
related. Accordingly, members which have many properties in common
with other members of that category are also likely to be rated as highly
prototypical. This might suggest that members located at the center of a
grammaticalization chain are likely to have the highest ratings for prototyp-
icality. It would seem, however, that more empirical data on grammaticali-
zation chains are required before generalizations of this nature can be for-
mulated, as we will see in Section 5.
To summarize, grammaticalization chains fulfil the criteria proposed
for family resemblance categories; they may be defined as family
resemblance categories of the linear type. This description, however, takes
care of only one aspect of their structure. It should be pointed out at this
stage that, rather than forming an explanatory parameter for grammaticali-
zation, family resemblance can be viewed as an outcome of grammaticaliza-
tion, resulting from the cognitive and pragmatic manipulation leading to the
emergence of new grammatical uses of erstwhile lexical forms (see Heine,
Claudi & Hunnemeyer 1991a). There are a number of other characteristics
which set grammaticalization chains off from other kinds of family
resemblance categories, in particular the following (see Section 2):
(g) They have a diachronic dimension, according to which the leftmost
member typically represents the earliest, and the rightmost member the
latest stage of development.
(h) Grammaticalization chains have a cognitive-semantic dimension,
according to which the leftmost member represents the most concrete con-
ceptual content, and the rightmost member the most abstract one.
(i) They also have a clear-cut morphosyntactic structure, according to
which the leftmost member has the highest and the rightmost member the
lowest degree of paradigmatic and syntagmatic variability (Lehmann 1982;
Heine & Reh 1984: 67).
That the example looked at above does not constitute an isolated
instance can be shown by looking at alternative cases of grammaticaliza-
tion, where the same basic structure can be observed. Table 3 presents an
abstract form of this structure.
GRAMMATICALIZATION CHAINS 351
Stage I II III
Meaning lexical grammatical grammatical
Form lexical lexical grammatical
Most research, both psychological and linguistic, has been based on the
assumption that family resemblance structures, wherever they can be
observed, are prototype categories, or can be described in terms of a pro-
totype framework (see especially Rosch & Mervis 1975). For example, both
prototypes and family resemblances involve context-dependent abductive
judgement (Givón 1989: 39), are based on a principle of relative similarity
and have fuzzy boundaries, and thus contrast with classical categories which
are based on the criterion of identity vs. non-identity and are defined in
terms of one or a few necessary and sufficient criterial properties.
In spite of such similarities, however, it would seem to be advisable to
keep prototypes and family resemblances apart in the case of categories
such as the one looked at here. This is also the conclusion reached by
Lichtenberk (1990) in his analysis of the grammaticalization of verbs of
motion ('go', 'come' and 'return') in Oceanic languages:
In the present study, the notion of a central or prototypical meaning or
function of an etymon is not relevant. The various grammatical functions
of each of the etyma discussed here belong in very different categories, and
the notion of prototypical functions does not arise. Thus it does not make
sense to ask whether, for example, the repetition-marking function or the
reflexive-marking function (both deriving from the meaning 'return') is
more central than the other. This is even true when comparing the source
verbal meaning and a derived grammatical function. The meaning 'return'
is primary historically, but at present it cannot be said to be more central
than, for example, the reflexive-marking function that developed from it
(Lichtenberk 1990: 76).
There are in fact a number of objections to analyzing examples such as
the one discussed in Section 2 in terms of a prototype framework. For
example, the member representing the "most basic meaning" in the study
of Heine, Claudi and Hunnemeyer (1991a), referred to above, is not a cen-
tral or "core" member of that category, rather it is a peripheral member,
being located at one endpoint of the relevant category. As we observed in
Section 3, however, it is central members of grammaticalization chains that
share the highest number of attributes with other category members, while
peripheral members have an extremely low number of category attributes.
Second, if indeed lexemes such as -yyéú were to be described in terms
of prototype theory, the question would arise as to whether all the various
members or stages indeed are part of one and the same prototype category.
GRAMMATICALIZATION CHAINS 353
may refer to, respectively, as discrete logic, prototype logic, and family
resemblance logic, and all three kinds of reasoning were constantly present
when our consultants described or defined their folk biological categories.
It would seem in fact that the relative extent to which these kinds of reason-
ing are employed in a given situation depends crucially on the relevant
nature of the "real world" phenomena considered. That Rosch and Mervis
(1975) observed a strong correlation between family resemblance and pro-
totype structures is to a large extent due to the fact that the categories cho-
sen by them were ones for which norms for the prototypicality of items had
already been obtained in previous research. 18 Categories having a pro-
nounced polycentric (or polysemous) structure, such as the one considered
in Section 2, do not figure in their sample.
6. Polygrammaticalization
ceived by the native speaker as belonging to one and the same linguistic cat-
egory. In the case of a more distant location, native speakers may no longer
be aware of any relationship between them and regard them as different
categories, as appears to be the case in English.
7. Polysemy
Thus, while there may be reasons for retaining the above set of criteria
for "canonical" instances of polysemy, there are also advantages in defining
polysemy with reference to general principles of grammaticalization
(Lehmann 1982; Heine & Reh 1984; Bybee & Pagliuca 1985; Heine, Claudi
& Hünnemeyer 1991a) and extending the use of the term polysemy to
grammaticalization chains like the one discussed in Section 2. This would
not only account for semantic variation but also for variation in the mor-
phosyntactic behavior and in the phonological shape of the words con-
cerned, as well as for the fact that semantic and morphosyntactic related-
ness in polysemy have both a synchronic and a diachronic dimension.
Note, however, that these observations are not of any help in deciding
where the boundary between polysemy and homonymy is to be located.
This problem would seem to require a different kind of approach, one that
takes, e.g., native speakers' intuitions on linguistic relatedness into consid-
eration.
8. Conclusions
As was mentioned above (Section 4), family resemblance does not pro-
vide an explanatory parameter for grammaticalization, rather it is an out-
come of grammaticalization, which itself is the result of an interplay of cog-
nitive and pragmatic forces leading to chain-like structures such as the one
sketched in Section 2 (see Heine, Claudi & Hünnemeyer 1991a).
There are a number of questions that have not been addressed in the
present paper. One of them concerns the conceptual nature of gram-
maticalization chains. For example, is Stage V of the -yyéú chain a concep-
tually impoverished version of Stage I, as proponents of the "bleaching
model" would argue (cf., e.g., Bybee & Pagliuca 1985), or can the develop-
ment from Stage I to V be described in terms of a "loss-and-gain model"
instead (Sweetser 1988)? Recent research suggests that while the latter
model provides a more appropriate account of grammaticalization than the
former (Heine 1991), the answer is more complex and requires a separate
treatment.
362 BERND HEINE
Author's address:
Bernd Heine
Institut für Afrikanistik
Universität zu Köln
D 5000 Köln 41
Germany
NOTES
* I wish to express my gratitude to Eithne Carlin, Ulrike Claudi, Colette Craig, Suzanne
Fleischman, Spike Gildea, Tom Givón, Nikolaus Himmelmann, Friederike Hünnemeyer,
Christa König, Andreas Lessau, Thomas Miiller-Bardey, Franz Potyka, Hans-Jürgen
Sasse, and Fritz Serzisko for valuable comments on an earlier version of the present
paper. I also wish to thank the Deutsche Forschungsgemeinschafi (German Research
Society), especially Frau Ursula Far-Hollender, for having sponsored the research on
which this paper is based.
1. In this paper the following abbreviations have been used:
ABS - absolutive case; F - feminine gender; INF - infinitive; M - masculine gender; NAR
- narrative tense; NEG - negation marker; NOM - nominative case; PAST - past tense;
PFV - perfective; PL - plural; QU - question; REFL - reflexive; SG - singular; 1 - first
person; 2 - second person; 3 - third person.
2. Concerning the distinction between "grammaticalization chains" and "channels of gram-
maticalization", see Heine, Claudi & Hiinnemeyer 1991a: Ch. 8.4.
3. Concerning the relation between cognitive and linguistic processes involved in categoriza-
tion, see Corrigan (1989).
4. The following tonemes are distinguished, which have both a lexical and a grammatical
function: high (e.g., á), low (a), and falling (a).
5. The function of the k- prefix in 1st and 3rd person verb forms in Charnus and other Maa
dialects is unclear. Its use in Charnus, though not in the Maasai dialect, is obligatory in
the imperfective and optional in the progressive and perfective aspects. The feminine (F)
gender prefix n- has the allomorphs m- before bilabial consonants and ŋ- before velar
consonants.
6. When -yyéú is followed by a verb, the latter is obligatorily in the "narrative tense".
7. It is possible to place the subject noun phrase sentence-initially, but in this case, the verb
is introduced in the form of a relative clause.
8. Concerning the distinction focal vs. non-focal sense, see Heine, Claudi & Hiinnemeyer
(1991a: Ch. 4).
9. In these contexts, the form (k)éyyeu tends to be preferred to (k)eyyéú. Note that the
function as a negation marker is the result of a specific invited inference which has not yet
been conventionalized to a distinct sense or "polyseme".
GRAMMATICALIZATION CHAINS 363
10. See especially Givón (1973; 1975 1979), Stephany (1982), Lehmann (1982), Heine & Reh
(1984), Bybee & Pagliuca (1985), Traugott & Heine (1991), Heine, Claudi & Hün-
nemeyer (1991a).
11. In Croft's treatment adjectives are considered in addition to nouns and verbs, and in addi-
tion to their discourse function, he is also concerned with their syntactic and semantic
functions.
12. Markings for number, gender, case, definiteness, etc., are said to be characteristic of
nouns, while markings for tense/aspect, modality, agreement, or the ability to take sub-
jects and objects, are said to be characteristic of verbs (cf. Croft 1984: 59-60; Hopper &
Thompson 1984; 1985).
13. Cf. Taylor (1988; 1989), who provides the following characterization of such categories,
involving the grammaticalization of adpositions: 'Tor each preposition, we recognize a
central, or prototypical sense. The prototypical sense, rather than being highly general,
may well profile a very specific configuration. Polysemy comes about when the preposi-
tion is used in a sense which is closely related to, but distinct from, the prototypical
instance. For example: a condition which is essential to the prototype might not be met;
a feature which is optional to the prototype now assumes central importance, or vice
versa; or some additional feature might be required. By the same process, this derived
meaning may in turn give rise to a further extension, and so on. The various senses of the
word thus radiate out from the central prototype, like the spokes of a wheel. Senses at the
periphery might well have little in common, either with each other, or with the central
sense; they are merely related by virtue of the intervening members of the meaning
chain" (Taylor 1988: 301).
14. This observation in itself may not be revealing, since our focus in the analysis was on dis-
tinguishing, rather than on identical, features between the various members.
15. One characteristic, namely that none of the stages combines all the attributes distin-
guished, applies only to extended chains such as the one presented above, but does not
apply to nuclear chains such as the one presented in Table 3, which have only three
stages.
16. Concerning the distinction macrostructure vs. microstructure, see Heine, Claudi & Hün-
nemeyer (1991a: Chapter 4).
17. "Family resemblances (even broadly defined) are undoubtedly not the only principle of
prototype formation — for example, the frequency of items and the salience of particular
attributes or particular members of the categories (perceptual, social, or memorial sali-
ence) as well as the as yet undefined gestalt properties of stimuli and stimulus combina-
tions, undoubtedly contribute to prototype formation ..." (Rosch & Mervis 1975: 599).
18. These categories include superordinates (furniture, vehicle, fruit, weapon, vegetable, and
clothing), basic-level categories (e.g., chair, car), or artificial categories consisting of
strings of letters (e.g., HPNWD, JXPHM).
19. For more details, see Westermann (1907), Hünnemeyer (1985). That le in (16) is not a
verb is suggested, inter alia, by the fact that verbs may not occur sentence-initially except
in singular imperative forms.
20. For more details, see Heine (1990).
21. Note that in many languages, including English (cf. Lyons 1977: 647), the distal
demonstrative ('that') is the unmarked member of the demonstrative paradigm.
364 BERND HEINE
22. As has been pointed out by Sweetser (1988) and Heine, Claudi and Hünnemeyer (1991a:
Ch. 4), however, in addition to the losses involved in this process, there are also gains, in
that, e.g., the loss of spatial semantics is compensated for by gains in text pragmatic sig-
nificance.
23. Criterion (c) tends to be implied rather than explicitly expressed in definitions of
polysemy (cf. Lyons 1977: 555).
REFERENCES