You are on page 1of 6

This article was downloaded by: [Arizona State University]

On: 16 October 2014, At: 16:01


Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Quantitative Linguistics


Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/njql20

Fractals in language
a
Ludˇk Hrebíček
a
Czech Academy of Sciences , Oriental Institute , Orientální
ústav AVCR, Pod vodórenskou věží 4, 182 08 Praha 8, Prague,
Czech Republic
Published online: 21 Jul 2008.

To cite this article: Ludˇk Hrebíček (1994) Fractals in language , Journal of Quantitative
Linguistics, 1:1, 82-86, DOI: 10.1080/09296179408590001

To link to this article: http://dx.doi.org/10.1080/09296179408590001

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever
as to the accuracy, completeness, or suitability for any purpose of the Content. Any
opinions and views expressed in this publication are the opinions and views of the
authors, and are not the views of or endorsed by Taylor & Francis. The accuracy
of the Content should not be relied upon and should be independently verified
with primary sources of information. Taylor and Francis shall not be liable for any
losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection
with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.
Terms & Conditions of access and use can be found at http://www.tandfonline.com/
page/terms-and-conditions
Journal of Quantitative Linguistics 0929-6174/94/0101-0082$6.00
1994, Vol. 1, No. 1, pp. 82-86 © Swets & Zeitlinger

Fractals in Language*
Luděk Hřebíček
Oriental Institute, Czech Academy of Sciences, Prague

ABSTRACT

The present paper is an attempt at the derivation of the Menzerath-Altmann law from the postulates of the
Mandelbrot theory of fractals. For this purpose the treatment of language system is characterized briefly, the
basic notions of the theory of fractals are described in an informal way, the sense and canonic derivation of the
Downloaded by [Arizona State University] at 16:01 16 October 2014

Menzerath-Altmann law is reproduced and, finally, the way of moving from the formula of the dimension of
self-similar fractals to this law is exhibited.

LANGUAGE AS A SYSTEM OF SUB- functioning as a educational discipline. The con-


SYSTEMS struction of a theory of language, then, seems to
be some sort of excess. It can be said that lin-
Linguistics differs from the other scientific disci- guistics as a science is in an evident crisis.
plines mainly in the unsystematic way of the This conclusion was already formulated twen-
construction of its edifice. This building looks ty years ago by Walter A. Koch in his foreword
like a construction knocked together out of in- to a book of collected papers presenting an in-
coherent parts, of cabins and annexes through troduction to linguistic theory:
which it is impossible to pass. Different lin-
guistic schools present their conceptions based 'Ohne daß sie es selbst eigentlich bemerkte,
mainly on definitions and classifications of lan- ist die Linguistik in eine Krise geraten. Die
guage constructs. This is analogical to the situ- Forschung ist aktiv wie nie zuvor. Und den-
ation in other sciences. However, linguistics noch droht die Gefahr der Paralyse, der Scho-
mostly refuses to build up scientific theories lastisierung. Das Bewußstwerden einer tief-
with reference to the alleged specific position greifenden Krise unter gleichzeitiger Mobi-
of natural languages among all other imagina- lisierung hoher wissenschaftlicher Energien
ble phenomena. Thus knowledge in general is kann andrerseits zu einer fruchtbaren Wende
divided into two parts which are language, on führen. ' (W.A. Koch 1973, p. XLVII)
the one hand, and all other items subjected to
cognition, on the other hand. Nevertheless, sci- This is an exact expression of the truth. Now
ence is a cultural phenomenon and it is neces- we should pose the question: In what way can a
sary to understand all cultural phenomena as a significant turn in the development of linguis-
whole. The science of language must be based tics be reached? This can only occur only when
on the same principles as the other sciences and two important conditions are fulfilled: (i) when
thus act as a part of the culture. linguistics tries to construct real scientific theo-
This situation of linguistics is partly due to ries based on the generally accepted epistemo-
the pragmatic demands put on linguistics (or logical principles, i.e. when each theory besides
better said, philology): for compilation of gram- the additional interpretative and conventional
mars, textbooks, manuals and dictionaries, for constituents (such as definitions, classifications,

* Correspondence should be addressed to: Dr. Luděk Hřebíček, Orientální ústav AVCR, Pod vodórenskou věží 4,
182 08 Praha 8, Czech Republic.
FRACTALS IN LANGUAGE 83

etc.) contains a core in the form of at least one mon with the mathematical meaning of this con-
testable (i.e. refutable) hypothesis; (ii) when lan- cept (being rather "coordination" or something
guage and its constructs are understood as items in this sense). Purpose and aim are not varia-
pertaining to one coherent system both in their bles; they are not sufficiently transparent cate-
mutual arrangements and in the process of their gories.
generation. The same situation occurs in some modern
Regarding the first condition: in a set of gen- linguistic denominations. The generative and
uinely theoretical works it was already convinc- transformational school treats certain language
ingly proven that quantitative linguistics is the levels as interpretative derivations of the main
only discipline able to construct advanced lin- chain of the chosen rules; in this way the pho-
guistic theories; see especially the works by G. nological and semantic levels are interpreted in
Altmann (1988,1992a), or by R.Köhler (1986). relation to the abstract generated structures. This
indicates that the mutual relation of these sub-
Downloaded by [Arizona State University] at 16:01 16 October 2014

The second condition may appear to be sur-


prising, because linguistics of the twentieth cen- systems is also "relative".
tury continuously stresses its intention to grasp Here we try to demonstrate that quantitative
language as a system. It can be indicated, how- linguistics developed one of the possible means
ever, that all schools of classical and modern by which linguistic theory can surpass the par-
linguistics explicitly treat language as a system ticularistic conception of the language subsys-
of relatively independent subsystems. And it is tems. We mean the Menzerath-Altmann law,
evident that the formulation of the relativity and discovered in 1980, see G. Altmann (1980). Up
independence of language subsystems is based to now a large set of works has been published
on the extent of our theoretical knowledge but dealing with this law from the viewpoint of its
not on the facts determined in languages. This theoretical structure as well as from the view-
peculiar relativity reflects our inability to pene- point of the testing of observed data, see G.
trate into the ties connecting the individual lan- Altmann & M.H. Schwibbe (1989), R. Köhler
guage subsystems. (1986), A. Fenk & G. Fenk-Oczlon (1993) and
In philological practice, the units of the indi- others. The Menzerath-Altmann law can be ap-
vidual linguistic levels are described independ- plied to different levels in different languages.
ently of the other levels. Phonemes, morphemes, This law can also be exploited for the purpose
lexemes, as well as syntactic constructions, are of the discovery of new, still unknown language
related to the semantic level; semantics is un- levels as there is no reason to suppose that the
derstood as a transparent phenomenon equally number of language levels is limited. The law's
understandable by anybody. Naturally, this is immensely wide range of applicability doubt-
not in agreement with reality. There are many less has some general reason. In what follows,
works trying to combine two levels as is the we try to explain this property through refer-
case of morphophonology or morphosyntax, etc. ence to the relation of the law to the theory of
These disciplines, too, are managed with the fractals.
help of definitions and classifications, but with-
out authentic theoretical means. In this connec-
tion it must be stressed that the highest lan- SOME FUNDAMENTAL CONCEPTS OF THE
guage level which theoretical linguistics is able FRACTAL THEORY
to take into account is the sentence level, and
this seems to be unsatisfactory. The theory of fractals is today generally known
The concept of function has been coined in and applied in many disciplines, but not in lin-
certain linguistic schools with an obvious in- guistics. The concept of 'fractal' was coined by
tention to subordinate many different language B.B. Mandelbrot during his investigation of cer-
phenomena to this notion. In such attempts, how- tain very complex and chaotic phenomena and
ever, 'function' means something synonymous processes. Fractal geometry is an ingenious dis-
to "purpose" or "aim" and has nothing in com- covery; it is suitable for the explanation of many
84 LUDËK HREBfÖEK

complicated phenomena which could otherwise line corresponding to the function


not be comprehended.
As for the concept of fractals, B.B. Mandel- N (8) = a 8"°.
brot (1982, p. 15) states the following:
The dimension D, in another example, repre-
'A fractal is by definition a set for which the sents an invariant of the scaling used by an ob-
Hausdorff-Besicovitch dimension strictly ex- server who instantly gets near a measured coast
ceeds the topological dimension. ' and the measure 8 limitlessly approaches zero.
Another model of a fractal set quoted by B.B.
This definition contains at least one concept for
Mandelbrot (and by other theoreticians of frac-
which a linguistic sense can hardly be found.
tal geometry) is the Koch curve. The construc-
Therefore we had better turn to the characteri-
tion of this curve starts from a straight line of
zation of the notion of fractal by J. Feder (1991,
length L(l) = 1. This line is called an initiator
Downloaded by [Arizona State University] at 16:01 16 October 2014

p. 19, sect. 2.3., we translate the author's for-


and in the second step of construction it can be
mulation from the Russian edition):
substituted by a generator and so forth, so that
'A structure consisting of parts, which are in each straight line of the unfolded formation is
a sense similar to the whole, is called "frac- substituted by the generator. The first three steps
tal". ' of the construction of the Koch curve are graph-
ically shown in the Figure 1.
A shorter characterization of fractal, by the same
author, is as follows:
'A fractal looks the same regardless of the INITIATOR

scale in which it is observed. '


This characterization indicates that in the sets
(structures, systems) comprising fractals the
property of similarity is involved in a certain
specific sense. When parts of a whole are simi- GENERATOR
lar to the whole (or vice versa), the system is
self-similar. Let us suppose a curve, the length
of which is measured in the number N(8) of the
straight lines (rods, paces, yard-sticks, etc.) of
length 8 covering the curve. The length of the
curve is L = N(S)8. This way of measuring is
complicated by the fact that L is function of 5.
This is usually exemplified by the measuring of
the coastline by a rod: if the rod shortens ad
Fig. 1. Koch curve.
infinitum (i.e., 8 -> 0), the length of the coast
grows infinitely.
Similarly, instead of a curve a surface can be This set evidently changes in the individual steps.
supposed being measured by the number N(8) Something, however, remains unchanged.
of squares 8 2 , or a volume N(8) by cubes 83, etc. B.B. Mandelbrot presents the following char-
For a very short 8 it can be proved that acteristics in connection with this curve:
N (8) ~ 1/8°. '... we assume that N "parts" into which our
"whole " is divided all involve the same sim-
If the number of squares 8 x 8 required for cov- ilarity ratio r. ' (B. Mandelbrot 1982, p. 56)
ering a coastline is N(8), the graph of the mutu-
al relation of log,0[N(8)] and Iog108 is a straight From Figure 1 it is evident what is denoted as
FRACTALS IN LANGUAGE 85

"whole" and "part" in the geometrical draft. For After its application to different data con-
the similarity dimension Mandelbrot presents cerning constituents and constructs of different
the following formula: languages, this law remains irrefutable, i.e. no
one has succeeded in refuting it - which has far
logiV reaching consequences. This law represents the
D = (3)
log (1/r) only objective criterion for differentiation of
language levels. As was already mentioned, there
When the respective data for the Koch curve is no reason to suppose that all levels are evi-
are substituted, the formula is as follows: dent at first sight and that their number is finite.
This is the case of the text units called 'aggre-
l0g4
D = (4) gates' (or, better said, 'sentence aggregates',
log (1/3) though both terms are provisional); see L. Hre-
Hence, it is evident what is meant by "parts" bîcek (1989, 1990, 1992a, 1992b, 1993a) and
Downloaded by [Arizona State University] at 16:01 16 October 2014

(N) and by similarity ratio (r) in the quoted C. Schwarz (1992). The other works analysing
passage. Dimension D thus appears to be the different spheres of application of the law are
characteristic of the sought-after invariant proper quoted above. This law also functions as an im-
to the mutable set. portant structure in the distribution of word as-
For the self-similar fractals, the Hausdorff- sociations; see G. Altmann (1992b) and L. Hre-
Besicovitch dimension D is identical to the sim- bfcek (1993b).
ilarity dimension D defined by formulae (1) and Why does this law have such a large range of
(2). application? This question deserves clarifica-
tion.
THE MENZERATH-ALTMANN LAW
A NEW WAY OF DERIVATION OF THE
This law was derived by Altmann on the basis MENZERATH-ALTMANN LAW
of the following differential equation:
The sought-after explanation can possibly be
y (5)
presented as a discrete way of derivation of the
y Menzerath-Altmann law. Let us return to the
Mandelbrot quotation concerning the Koch curve
withy = length of the language construct, which was presented in the preceding section.
x = length of the respective constituent, N "parts" of a "whole" involve the same simi-
b = coefficient. larity ratio r. When these two variables are con-
The solution of (5) results in fronted with the two variables of the Menzerath-
Altmann law, it is obvious that they are parallel
log y = b log x + c in their meanings: the construct of length x (or
N) consists of y (or r) "parts". Formula (1) then
and, consequently: can be rewritten as

y = Ax» (6) log*


D = (7)
log (1/y) -logy
c
with A = e .
Then we obtain:
The Menzerath-Altmann law asserts that
logy = - ( I / o ) log* (8)
the longer the language construct the short-
er its constituents. This expression can be supplemented by the
correction term A in its logarithmic form. Fur-
It is obvious that formula (6) holds only for b < thermore, b can be substituted for 1/D. The equa-
0. tion:
86 LUDËK H R E B Î C E K

log y = - b log x + log A (9) constant or whether it varies as a function of an


as yet unknown argument. Nevertheless, thanks
is identical with (6), i.e. with the formula of the to the discovery of the Menzerath-Altmann law,
Menzerath-Altmann law. linguistics has found its own route to fractals.
This situation can be interpreted in two ways:
1. The derivation of the Menzerath-Altmann law
represents nothing more than a structural ho- REFERENCES
mogeneity of the two formulae.
2. The identities N = x and r s y, supposed Altmann, G. (1980). Prolegomena to Menzerath's Law.
Glottometrika, 2, 1-10.
above, are correct; they are in agreement with Altmann, G. (1988). Wiederholungen in Texten. Bochum,
reality and thus the derived law is a real conse- Brockmeyer.
quence of the fractal structures existing in lan- Altmann, G. (1992a). Science and Linguistics. In: R.
guage. Köhler & B.B. Rieger (1993), 3-10.
Downloaded by [Arizona State University] at 16:01 16 October 2014

Altmann, G. (1992b). Two Models for Word Associa-


If the second variant is correct, and every- tion Data. Glottometrika, 13, 105-120.
thing testifies that it really is, then the deriva- Altmann, G., & Schwibbe, M.H. (1989). Das Menzerath-
tion presented above can be taken as an expla- sche Gesetz in informations verarbeitenden Syste-
nation of this law's wide range of application. men. Mit Beiträgen von Werner Kaumanns, Rein-
hard Köhler und Joachim Wilde. Hildesheim-Zürich-
In addition, we should stress the following facts: New York, 01ms.
the Menzerath-Altmann law holds for all possi- Feder, J. (1988). Fractals. New York, Plenum Press.
ble language constructs and their constituents; [Russian edition (1991): Fraktaly. Moskva, Mir.]
to be a language construct means, therefore, to Fenk, A., & Fenk-Oczlon, G. (1993). Menzerath's Law
and the Constant Flow of Linguistic Information. In:
have constituents in the sense of the law, and R. Köhler & B.B. Rieger (1993), 11-31.
thus constructs and their constituents represent Hřebíček, L. (1989). Menzerath-Altmann's Law on the
two different levels (or subsystems); the Men- Semantic Level. Glottometrika, 11, 47-46.
zerath-Altmann law is derivable from the ex- Hřebíček, L. (1990). The Constants of the Menzerath-
Altmann Law. Glottometrika, 12, 61-71.
pression valid for self-similar fractals. Hřebíček, L. (1992a). Text in Communication: Supra-
Future investigation should decide whether Sentence Structures. Bochum, Brockmeyer.
the language system is really self-similar or Hřebíček, L. (1992b). The Relation "to Consist of. In:
I. Prosecký (ed.): Ex Pede Pontis. Prague, Oriental
whether it is affine conglomerate and its self- Institute (AS), 80-85.
similarity is nothing but a consequence of the Hřebíček, L. (1993a). Text as a Construct of Aggrega-
statistical approach to the measurement of the tions. In: R. Köhler & B.B. Rieger (1993), 33-39.
length of constituents; variable y represents a Hřebíček, L. (1993b). Word Associations and Text. Glot-
tometrika, 15, 12-17.
mean value characterizing the entire distribu- Koch, W.A. (1973). Einleitung. In: W.A. Koch (ed.):
tion of data displaying the length of constitu- Perspektiven der Linguistik I. Stuttgart, Kröner, XI-
ents. The fractal dimension D = 1/b, then, rep- LV.
resents an invariant of the process of approach- Köhler, R. (1986). Zur linguistischen Synergetik: Struk-
tur und Dynamik der Lexik. Bochum, Brockmeyer.
ing from one level to the other one (upward or Köhler, R., & Rieger, B.B. (1993). Contributions to
downward) similar to the process in which an Quantitative Linguistics. Dordrecht-Boston-London,
observer approaches the measured coastline or Kluwer.
when the construction of the Koch curve pro- Mandelbrot, B. (1982). The Fractal Geometry of Na-
ture. New York, Freeman.
ceeds step by step. The dimension D expresses Schwarz, C. (1992). Zur Verteilung von Aggregaten in
something which remains fixed during those Texten. [Unpublished seminar work.] Bochum,
changes. Future investigation of languages should Sprachwissenschaftliches Institut der Universität.
decide whether D (or its reversed value b) is

You might also like