Professional Documents
Culture Documents
Nikolaus Ritt
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Contents
1 Introduction 1
2 The historical perspective 11
3 Approaching ‘language change’ 19
4 The Darwinian approach 62
5 Generalising Darwinism 89
6 Towards an evolutionary theory of language 122
7 What does all this imply for the study of language
change? 230
8 How to live with feet, if one happens to be
a morph-meme 240
9 The prosodic evolution of English word forms or
The Great Trochaic Conspiracy 289
10 Conclusion 307
References 313
Index 323
v
Figures
vi
List of figures vii
ix
x Preface
1
2 Selfish Sounds and Linguistic Evolution
Most human societies are aware of the limitations of and the dangers
inherent in human language and many have tried to come up with ways
of reducing language related risks: children are sent to school, are trained
in the most profitable use of their mother tongues, and are taught to
understand and see through the rhetorical tricks of demagogues. Also,
a considerable and growing number of people all over the world are
taught foreign languages, so that the gaps between linguistically differ-
ent communities are more easily bridged. Finally, institutions of research
and higher education that dedicate themselves to the study of human
language have spread all over the globe during the last one hundred
years.
tries to work out how they interact to produce its specific effects. If one
succeeds, one can then reconstruct the plan and the intentions behind
the tool. Unfortunately, however, this approach faces serious problems
when applied to language, since many of its aspects are simply impossible
to dismantle in such a way that their constituents could be easily isolated
and observed. This has several reasons. To begin with, it is not at all
obvious what exactly to take apart if one wants to lay open ‘the internal
mechanics of language’. Language seems to manifest itself in a variety of
different domains, such as in texts, in behaviour, in individual speakers’
competence, or in social conventions. Where exactly, and in what manner
does ‘it’, that is, the tool that we are interested in, exist then? Which, if any,
of its manifestations should be considered primary? As we shall see below,
the issue is rather complex and forces one to make subtle, yet principled
decisions.1 Secondly – albeit closely related to the ontological problem –
there exist good reasons to suspect that at least much of language is part
of the human mind. The mind, of course, is still somewhat of a white
spot on the scientific landscape, and relatively little is known about it. To
make things worse, all that is known about it suggests that the properties
of minds depend most crucially on the workings of human brains, and –
for both practical and ethical reasons – we are in no position to take those
apart for the purposes of academic enquiry.2
Now, if one cannot dismantle a tool and look at its parts, the only way
in which one can try and develop an idea about its internal mechanics is
through inference. One observes the behaviour of the tool under variable
and controlled conditions and then tries to imagine what kind of con-
struction could achieve the observed effects. The hypothetical blueprint
which one thus constructs might also be called a ‘theory’ or ‘model’ of
the tool. The problem with such indirectly derived models is that one
can never be sure how similar they actually are to the ‘real’ machine
of which they are supposed to be models. One will never really know
if the model and the original look alike inside, even though both might
‘behave’ almost identically. For practical purposes, this may not make a
1 It might be necessary to stress already at this point, however, that ‘language’, if it is viewed
as a tool, cannot at the same time be identified with ‘texts’. Text, i.e. the output produced
with language tool, can of course be ‘taken apart’ and analysed rather easily (at least in
certain ways), but the same does clearly not hold for the ‘tool’ itself, which includes the
mental machinery involved in both producing and understanding textual output.
2 As a matter of fact, the last decades have seen the development of techniques by which
activity within human brains can actually be measured and recorded without damaging
the brains themselves. The best known ones are ‘Positron Emission Tomography’ (PET),
‘Magnetic Resonance Imaging’ (MRI) and the ‘Superconducting Quantum Interference
Device’ (SQUID) (Rose 1992: 131–4). It is fair to say, however, that the measurements
they permit are still fairly rough and don’t yield sufficiently fine-grained pictures for most
linguistic purposes, so that the claim to which this footnote refers is still largely valid.
Introduction 5
big difference. Having a good ‘model’ might even put one in a position
to design new tools that are just as efficient as, or even better than, the
original in performing certain tasks. Still, one will never ultimately know
whether one’s model faithfully represents the internal make-up of the
original, and there will always remain the possibility that circumstances
might arise – not encountered before – in which one’s model will behave
differently from the original after all. Should this happen, one will have
to revise and adapt one’s model accordingly.3 In short, modelling a tool
through inferring its internal mechanics from its observable effects tends
to strike one as somewhat unsatisfactory, yet if the tool under considera-
tion is language, it is the only choice one has.
textual shapes in which it comes. For example, the word language does
not only have a shape, but also expresses some meaning and it is only
this that makes it language. Otherwise it would just be a pattern of black
shapes on white background. In order for textual patterns to ‘have mean-
ing’, however, their form is not sufficient. Instead, speakers need to be
involved: either those who produce texts to ‘express’ meaning, or those
who interpret them to ‘recover’ it. It is important to note that the kind
of meaning that gets associated with particular textual patterns depends
at least as much on what speakers do with them as on the structures of
the texts themselves. The American philosopher Daniel Dennett (1990,
and http://ase.tufts.edu/cogstud/papers/intrptxt.htm) has contrived a nice
little text which will get two radically different meanings when ‘pro-
cessed’ by either English or French speakers, and which illustrates the
often underestimated role which speakers play in endowing texts with
meaning.
(1) –
In the first case, it can be interpreted to ‘mean’ or to ‘express’ something
like You have (a) great leg(s). Why don’t you touch ours/mine?, in the second
something like Great heritage! Sixteen bears! Of course, this example is
made up for the purpose and not a very natural text, but it does drive the
point home quite impressively. Language must be more than texts.
From a different perspective ‘language’ could, for instance, be regarded
as a form of human behaviour that involves mental and physiological
processes somehow linking ‘meanings’ to ‘texts’. An established term for
language in this dynamic, procedural sense is ‘discourse’.4 Observing
and describing it is more challenging in many ways than analysing static
texts, but both the physiological aspects of the processes involved (such
as articulation or auditory perception) as well as the behavioural context
of discourse (including many of the effects it has on people, for example)
are still relatively amenable to detached, empirical observation.
Yet, even communicative behaviour cannot be all there is to language.
After all, there is a sense of ‘language’ in which speakers ‘have it’ even
while they do not actively use it. It appears to exist in speakers’ minds
as a cognitive potential for producing or interpreting an infinite number
of possible utterances. Often referred to as linguistic ‘competence’ (e.g.
Chomsky 1965: 4), language in this sense represents a system of knowl-
edge which speakers draw upon when they engage in linguistic behaviour
and produce or interpret texts. This cognitive or mental implementation
4 Defined in Beaugrande 1997 as ‘the level of the total communicative event, including
discoursal moves, gestures, facial expressions, emotional displays, and so on, in contexts
of situation’ (44).
Introduction 7
5 ‘For language [langue] is not complete in any speaker; it exists perfectly only within a
collectivity’ (Saussure 1959: 21).
8 Selfish Sounds and Linguistic Evolution
gets so attached to one’s perspective that one becomes all but incapable of
questioning its basis. Certainly, this may be socially safe. If one adopts a
theory that is shared by a substantial number of colleagues, one can count
on their goodwill even if only for joining their ranks. But should one’s
chosen approach be inherently flawed, one is unlikely to discover it that
way. Therefore, I have decided quite deliberately to approach my subject
matter as naively as possible. Risking reinventing one or the other wheel,
I shall try to describe the motivation of the present study, the particular
problems it addresses, the perspective it takes, and the assumptions it
makes in considerable detail in the following sections. I will be pleased if
I persuade some of my readers to follow me back to basics. Since I have
no intention of ‘impressing’ them, or ‘persuading’ them of the ‘ultimate
correctness’ of my argumentation, I shall try to make it as easy as possible
for them to take issue with the points I make. My hope is that they will
do so, detect all the flaws I am certain to have overlooked, and make the
best of it.
2 The historical perspective
11
12 Selfish Sounds and Linguistic Evolution
(3)
a. Temporal Adjunct Object Verb Subject
[not many dayes after] [the yonger sonne] [gathered] [al] [together]
Finally, the texts also differ in a number of other interesting ways. For
instance, text (2g) has the son turn ‘the whole of his share into cash’
before wasting it all, while in all earlier versions nothing specific is said
about the form in which the son decided to take his share of the family
property with him. Interestingly, also, in texts (2a) and (b), it seems to be
the elder of the two sons who asks the father to divide his property before
his younger brother then decides to leave home with his share and waste
it by living luxuriously, whereas in all the other versions it is the younger
son himself who asks his father to do so. A difference which changes the
moral of the whole story considerably.1
The textual differences mentioned in the last paragraph seem primar-
ily to reflect differences among the cultures within which they were pro-
duced, not among the languages in which they were written. Thus, it
has become normal in twenty-first century advanced capitalist societies
to convert one’s property into money or an even more virtual equivalent
if one wants to spend or, indeed, waste it conveniently, and this may
account for the innovative phrase turned the whole of his share into cash
in (2g). Similarly, the other difference might reflect different concepts of
divine justice held by the respective author-interpreters. To the authors
of texts (2a) and (b) it might have been inconceivable that God should
reward a sinner unless that sinner can in a way be exculpated – in this
case through an elder brother whose impatience to come into his prop-
erty forces his younger sibling to face a temptation before he is mature
enough to overcome it.
All other differences, however, that is graphic, phonological, inflec-
tional, or syntactic ones, are typically thought to reflect ‘changes in the
English language’ itself. This assumption is basically plausible, and rests
on the following type of reasoning. First, the differences among the texts
are taken to reflect parallel differences among the ‘languages’ in which the
texts were written – that is, differences on the level of individual or socially
distributed linguistic competence. Second, the texts (and by implication
the ‘languages’ they ‘are in’) are not merely different, but display more
similarities and correspondences among each other than one could expect
1 The story continues like this. When the son has spent his property and ends up having
to beg for food, he decides to go home to his father and ask his forgiveness. He does so
and his father not only forgives him but actually celebrates his homecoming in a big way.
This irritates the brother who has stayed home dutifully all the while, and he complains
that his own obedience ought to deserve a celebration much better than the homecoming
of his brother after years spent in luxury and sin. The father, however, replies that he
regards the repentance of a sinner as a happier occasion than the simple righteousness of
the righteous. Arguably, the father’s behaviour towards his older and more dutiful son is
a bit unfair and might be easier to understand if he had somehow been responsible for
the misfortune of his younger brother.
The historical perspective 15
3.1 Preliminaries
The words ‘language’ and the compound ‘language change’ are familiar
from everyday use. There they carry meanings and associations which
work well enough for everyday purposes. In a way, these meanings and
associations might be viewed as mostly implicit and rather crude theories,
or working hypotheses. All humans who are confronted by language in
their daily experience have them. Investigating the nature and mechanics
of language and language change, however, is definitely not a typical
everyday purpose. Approaching a language as something that changes and
has a history is clearly different from approaching language as something
that one learns, knows, uses or understands. Since the shape of any theory
will reflect the purposes for which it is constructed, it cannot be taken
for granted that the normal, everyday way in which we think of languages
should prove useful when we intend to understand their historicity. After
all, it is obvious that common sense often conceptualises phenomena
in a way which can be utterly inadequate for special, and in particular
scientific purposes. Take such concepts as sunrise and sunset, to give a
trivial but telling example. While they are perfectly adequate for referring
to the phenomena as most of us experience them, they are downright
deceptive for the purpose of describing or even understanding the celestial
mechanics behind the events. This means that although we may have no
other choice but to work with the established notions that constitute our
‘common sense’ when approaching the phenomenon of language and its
change, we must accept that our investigations might eventually cause
us to revise our understanding of the concepts, or even to give them up
altogether, like – to cite another well-known example – the notion that
there was one particular quality or substance, distinguishing life from
dead matter, had to be discarded together with the concept of élan vital.1
1 A term which was originally coined by the French philosopher Henri Bergson (1859–
1941) in 1907 in his book L’évolution créatrice. He proposed that what distinguished life
19
20 Selfish Sounds and Linguistic Evolution
The following sections will attempt to show that our everyday under-
standing of language is highly ambivalent and open to many different
interpretations. These will be made explicit, and it will be discussed
how the different phenomena which can be referred to as language hang
together and what roles they can play in an attempt to understand lan-
guage. This exercise should help us to avoid common sense notions which
might tacitly bias our understanding of linguistic change and thus prevent
us from conceptualising in the manner which is most appropriate to our
task.
from mere matter was an essence with which the former was imbued and which was so
ephemeral that science had not yet discovered it. He turned out to be wrong. See also
Russell (1961: 756–66).
2 There are understandings of the word ‘text’ which differ markedly from the definition
above and which are employed by text linguists or literary scholars. I am of course perfectly
aware of this. For the purposes of the present argumentation, and throughout most of this
book, however, I shall use the term in the straightforwardly materialist sense employed
here.
3 In spite of their ultimately neuro-biological character, the processes involved in linguistic
behaviour can also be described in terms of such higher-level concepts as those used
in psychology, for example. The question of how they are best described need not be
addressed at this point, however. For a debate of neuronally reductionist explanations
of cognitive and behavioural processes see for example Hubel (1988), Edelman (1989),
Fodor (1989), Zeki (1993), P. S. Churchland (1986) P. M. Churchland (1995), Crick
(1995), Snyder (1986), Chalmers (1996), McGinn (1999), Gold/Stoljar (1999), Chater
(1999), or Jamieson (1999).
4 Just how autonomous ‘linguistic competence’ might be from other parts of human neuro-
biological systems or to what degree it should be regarded as separate from them at all is
a non-trivial question and no rash decisions ought to be taken on it. Furthermore, and
just like in the case of linguistic behaviour, the neuro-physiological implementation of
Approaching ‘language change’ 21
linguistic competence should not be interpreted to rule out the possibility that its prop-
erties might be more suitably described in terms of higher-level concepts. Their essentially
material nature will be regarded as beyond reasonable doubt, however.
5 Of course, the question whether languages of which no textual evidence whatsoever nor
any second hand witnesses exist should also be assumed to ‘exist’ in this abstract sense is
more of philosophical than of practical relevance. If it is answered positively, however, I
see no principled way of denying ‘existence’ of the same kind to all future languages, or
even to all possible languages, which creates the uncomfortable situation that one would
have to speak of the ‘reality’ of the ‘unrealised’. For my taste this is bordering too closely
on the paradox to be rationally discussed.
22 Selfish Sounds and Linguistic Evolution
TEXTS
trigger
result
F in
THEORIES,
IDEAS, B
ABSTRACT INSTANCES OF LINGUISTIC
KNOWLEDGE ‘of’ BEHAVIOUR
(productive and/or receptive)
informs
shapes
informs
constrains
LANGUAGE
CAPACITY
INDIVIDUAL (biological basis
LINGUISTIC of competence
COMPETENCE and behaviour)
C
D
POOL/NETWORK of
consists of / COMPETENCES
emerges from within a
COMMUNITY
E
6 There are notable exceptions, of course. One of the earlier ones would be Hermann Paul’s
Prinzipien der Sprachgeschichte (1880), or Weinreich, Herzog and Labov’s seminal 1968
paper ‘Empirical foundations for a theory of language change’. That the latter explicitly
addresses and revises Paul’s views indicates that the discussion was not very intense in
between, however. See also McMahon (1994: 7f).
Approaching ‘language change’ 23
for describing and explaining language change. All one can possibly do
on the textual level alone is to identify counterparts and then chart corre-
spondences between them – that is, similarities and differences.7 A worth-
while and necessary endeavour, to be sure, but clearly only the starting
point for a serious investigation of the processes that actually constitute
linguistic change.
7 That texts do not actively undergo language change has always been recognised by lan-
guage historians. Already Hermann Paul, for example, stressed that
Das wirklich Gesprochene hat gar keine Entwicklung. Es ist eine irreführende Ausdrucks-
weise, wenn man sagt, dass ein Wort aus einem in einer früheren Zeit gesprochenen
Worte entstanden sei. Als physiologisch-physikalisches Produkt geht das Wort spurlos
unter, nachdem die dabei in Bewegung gesetzten Körper wieder zur Ruhe gekommen
sind. (1920: 28)
Approaching ‘language change’ 25
b. changes into
T1 T2
informs B alters C informs B informs
c. changes into
B1 B2
produces T informs B alters C informs
In practice, this is not the case, however. The reason is related to the
fact that we shall never have direct empirical access to all the individual
factors which underlie differences between any two stages of C-, B- or
T-language, irrespectively of how we look at it. Instead we shall always
have to make a large number of assumptions. If, for example, we regard
language change as something which happens to C-language and is in
principle ‘brought about’ through the mediation of linguistic behaviour
and textual output, we shall never be able to collect the necessary evidence
of all the actual instances of language behaviour nor of all the written and
spoken texts which contributed to bringing the differences between C1
and C2 about. All we can do is hypothesise about the kinds of linguistic
behaviour and the kinds of texts that must have been involved. Of course,
our hypotheses might achieve a respectable degree of plausibility, since
we can of course sample actual texts or instances of linguistic behaviour
and check whether they are at all compatible with our assumptions. Yet,
they will still remain hypotheses for all that.
If we ‘start’ on any of the other two levels, things would not seem to
be different at first. When we compare two temporally distant texts, for
instance, we will hardly ever know through what actual chain of events
they are linked and will also have to make assumptions about likely event
types instead. Yet, there seems to be one crucial difference between com-
petence on the one hand, and the other two manifestations of language
on the other. While the latter are typically strongly co-determined by
extra-linguistic factors such as the situation, the physical condition, the
mood, the social setting or the particular communicative intentions of
individual speakers, the former is not – at least not to a similar degree.
This follows from a simple observation: whatever the contingencies of
your upbringing, whatever texts you actually come across as a child
and whatever people say to you, you will wind up being competent in
English, as long as that is the language which is around for you to acquire.
Approaching ‘language change’ 29
On the other hand, no matter what your competence is like: if, when and
where you say anything, and what you say, does not only depend on your
knowing English but on a large number of other factors, which may fall
outside the linguistic domain proper.
In other words, differences between texts or instances of linguistic
behaviour may always have a variety of reasons which fall outside the
chain of causal links represented in (5). Furthermore, many of these
extra-linguistic factors are likely to be historically contingent, and this
makes stories about how texts or instances of linguistic behaviour are
causally linked difficult to tell – not exclusively, but predominantly from
a linguistic point of view. In a very crucial respect, then, things are not
quite the same on the level of competences. Because if it is true that
children can acquire the ‘same’ language from different sorts of textual
input as long as it is ‘in’ that particular language, this implies that in the
acquisition of linguistic competence many of the contingencies that make
texts difficult to compare with one another will be filtered out.
Although it may not at all be clear what informational input comes to
be treated as linguistically relevant in language acquisition and how the
developing minds of children manage to distinguish it from accidental
information, the conclusion that some distinction of that kind must be
made is unavoidable. Language acquisition can therefore be regarded as
a kind of sieve through which only linguistically relevant information can
pass, and we can revise the causal chain of events sketched in (5) at least
in principle as follows:
(6) X2 Y2 Z2 A2
X1 Y1 Z1 A1
X3 Y3 Z3 A3
T1 informs Bi alters C informs Bp informs T2
X4 Y4 Z4 A4
X5 Y5 Z5 A5
9 There is a way in which this statement is only partly true. As computer-aided studies of
text corpora have shown, there exist collocations and phrases in every language which
tend to recur quite frequently in individual utterances (see, for example, Sinclair (1992)
and Svartvik (1992)). These, albeit fairly small, stretches of text do represent formally
comparable counterparts of each other. As stretches of texts get longer, however, the
number of structural disparities among them rises drastically.
Approaching ‘language change’ 31
12 Thus using Chomsky’s term (for which see Chomsky (1986: ch. 2) or Maher/Groves
(1996: 15–19) slightly differently than he would probably say he does, but in accordance
with some critical interpretations of his way of using it.
34 Selfish Sounds and Linguistic Evolution
13 http://info.ox.ac.uk/bnc/(January 2000).
14 http://titania.cobuild.collins.co.uk/boe info.html (January 2000).
15 See, for example, de Beaugrande (1991: 147–87), or, Weinreich/Labov/Herzog (1968).
Approaching ‘language change’ 35
16 Actually, Paul’s statement is not completely adequate in its extreme form. Abstrac-
tions can interact causally, though only via their psychological realisations in human
brains. Furthermore, they may be logically connected, if they are parts of abstract formal
systems.
36 Selfish Sounds and Linguistic Evolution
3.3.7 Summary
Let us see what the arguments made so far imply for the interpretation of
statements such as ‘Old English’ has ‘changed into’ or ‘become’ ‘Present
Day English’.
What is usually called ‘Old English’ represents a heterogeneous (yet
most probably inherently ordered) pool of competences ‘in’ Old English.
These competences will each have been different from one another, but
will have shared a sufficient number of properties for making commu-
nication among ‘Old English’ speakers possible. ‘Present Day English’
represents another pool of competences, once again heterogeneous in an
orderly way. Importantly, the mix of competence properties that charac-
terises the pool constituting ‘Present Day English’ differs considerably
from the mix of properties that characterises the ‘Old English’ compe-
tence pool. Some properties that can be found in one pool are absent in
the other, and of those which are present in both some will be distributed
differently. We assume that these differences are due to ‘language change’.
This implies that some causal link can be established between the ‘Old
English’ competence pool on the one hand and the ‘Present Day English’
pool on the other. Most probably, such a link is established via behavioural
and textual manifestations of language, as well as by competences of
‘intermediate’ stages of English. It is supposed that temporally later com-
petences assume their characteristic properties by interpreting the textual
output produced on the basis of earlier competences. Linguistic change
happens because later competences do not always appear to assume quite
the same properties as earlier competences. Thereby, the mix of prop-
erties that characterises the competence pool of a speech community is
continually altered – albeit only slightly – as one new competence after the
other assumes first its adult, and ultimately its final state. At the same time
earlier competences are continually removed from the pool as the speak-
ers who host them die. Over time, these processes may amount to such
differences as those which distinguish ‘Present Day English’ from ‘Old
English’. This, then, is what we refer to when we say that ‘Old English’
38 Selfish Sounds and Linguistic Evolution
Stage 1
T1 C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 ... Cn
C1’ C2’ C3’ Cd C5’ C6’ C7’ C8’ Ci C10’ Ckt C12’ C13’ C14’ ... Cn
C1’‘ C2’‘ C3’‘ Cd C5’‘ Cf C7’‘ C8’‘ Ci C10’‘ Ck C12’‘ C13’‘ Cn ... Cn
Stage 2
T2 Ca Cb Cc Cd Ce Cf Cg Ch Ci Cj Ck Cl Cm Cn ... Cx
Legend:
C1 Old competence Cn Obsolescent competence Ca New competence
C1’ Altered competence Tn Text Bn Instance of linguistic behaviour
has become, or changed into ‘Present Day English’. The processes are
schematically represented in figure 3.2, which summarises, how the dif-
ferent levels on which language manifests itself interact to bring linguistic
change about.
A few things need to be stressed. First, and in spite of its apparent
complexity – the schema still represents only a very general framework
Approaching ‘language change’ 39
with long /e /s. Generalising over the two words, one can say that a cor-
respondence seems to hold between present day /i /s and earlier /e /s.
Interestingly, the correspondence holds in very much the same way when
one considers other words which are today pronounced with /i /s and
which had counterparts in earlier stages of English, such as the verbs
meet (OE mētan) and see (OE sēon), or the adjective green (OE grēne). In
fact, it is so general that it justifies a correspondence rule such as OE/ME
/e / ←→ ModE /i /.
Now, that the histories of all languages abound with regular correspon-
dences of this kind came to be established as fact as early as the eighteenth
century and gave rise to the research programme known as historical com-
parative linguistics, advanced most successfully during the nineteenth
century by a group of scholars commonly referred to as ‘Neogrammari-
ans’. It has admitted the writing of law-based sound histories and inspired
important conclusions about linguistic phonology. That the /e /s in words
so different as he, metan, or greene were apparently all affected by the same
type of change may be taken as evidence, for example, that all these /e /s
must indeed be instantiations of one and the same sound type: a point
which is not as self-evident as it might appear.
But all this need not concern us at the moment. Let us return to
the particular ‘change’ we are presently considering and which we sup-
pose to have been behind the correspondence between /e /s and /i /s
in words like he, me, see, meet and green, and their historical counter-
parts. In the community of historical English linguists, the ‘change’ is
well known. It is one in a set of changes which are normally referred to
as the ‘Great Vowel Shift’.20 These changes are likely to have occurred
at the end of the Middle or the beginning of the Modern English
period, although there is considerable disagreement about its exact
timing.
Now, for our purposes it is crucial to be extremely careful as we
approach evidence of a change in order to reflect upon the processes that
may have actually brought it about. For instance, we ought to remind our-
selves that the homogenous systems which words such as ‘Old English’,
‘Middle English’ or ‘Modern English’ suggest are unlikely to be histori-
cally real. Also, we try to distinguish sharply between the textual evidence
that the correspondence between /e / and /i / represents, and the entities
and processes we interpret it to be evidence of. In short, we must pre-
vent our common sense from performing generalisations and abstractions
which might turn out to be inadequate for historical linguistic purposes,
and resist the natural impulse to assume that there was a language called
Middle English at a particular time, that this language had an /e /, and
that this /e :/ ‘became’ /i / as Middle English ‘became’ modern.
What, then, should we allow the regular correspondence between /e /s
in Old English words, and /i/s in Modern English words to tell us? First
of all, we have to be aware that the words we are looking at do not rep-
resent first-hand data, but interpretations of such data. The first-hand
data on which they are based, are, first, contemporary pronunciations
of the words he, me, see and meet, and their graphic representations,
and, second, historical spellings of pre-sixteenth-century counterparts
of those words, passed down to us in the form of written documents.
These days, such historical documents are made accessible to the larger
academic community in the form of printed, or digitised editions of such
documents. Sometimes, selections of texts from various periods are col-
lected in computer corpora. Most famous among historical corpora of
English is the Helsinki Corpus.21 In it, we will find text passages like the
following.
(9) Trust veryly in God and leve hym and serve hym, and he wyl not
deseve w.
( 225: Heading)
Ther sall a childe borne be,
Goddis sone of heuen is hee
And man ay mast of myght.
( 124: Heading)
Alle þis worlde is wrothe with mee,
fi is wote I wele.
( 72: Heading)
Notwithstanding, by mine advice, if ye have this letter or the messenger
come to you, come to the Kinge wards or ye meete with him, and when
ye come ye must be suer of a great excuse.
( 201: Heading)
For him wes loþ men to mete;
Him were leuere meten one hen,
þen half anoundred wimmen.
(/2 25: Heading)
21 This widely known corpus was compiled at the University of Helsinki, under the direc-
tion of Matti Rissanen. It is part of a corpus collection available from the Interna-
tional Computer Archive of Modern and Medieval English. For further information see
http://helmer.asksis.uib.no/icame/newcd.htm.
42 Selfish Sounds and Linguistic Evolution
Tak þe grene bowes of an asche & bryne þam & kepe þe jeuse þat
commes owte at þe endis
(/4 6: Heading)
And the erthe broute forth greene erbe [. . .]
(3 ,1: Heading)
ther shaltow seen anoon thilke verray blisfulnesse that I have behyght
the.
(3 430.1: Heading)
first doe it, and be sen meke, that other may lere for to ouercome pride,
(2/4 25: Heading)
The stories which can be derived from data like these all require a
fair amount of conjecture. Thus, although it is very plausible, even the
belief that the sounds which the historical e(e)-spellings represented
were phonetically similar to [e ], depends on an interpretation of wit-
nesses. And the same is true, to a possibly even greater extent, with regard
to the mental properties assumedly underlying and expressed by e(e)-
spellings. There, however, the issue is even more complicated because
there are rather fundamental disagreements within the linguistic commu-
nity about how phonological competence should be conceived of at all.
It would therefore be clearly rash, though tempting, to take it for granted
that the fact that words like he, me, see, green or meet were spelt with e
or ee means that they were mentally represented by ‘the phoneme /e /’.
Yet, certain assumptions clearly have to be made for all the problems that
the reconstruction of historical pronunciations and phonologies poses,22
otherwise it is impossible even to start thinking about linguistic evolu-
tion. Thus, the following preliminary description strikes me as relatively
plausible and safe.
First, before the sixteenth century there seem indeed to have existed
counterparts of the ModE words he, me, see, green, meet, and so on; sec-
ond, their pronunciations are very likely to have involved [e ]-like sounds;
and third, phonological competences will have existed which are likely
to have had a property in common, namely that of which the graphic
e(e)s and the assumed phonetic [e ]s can count as ‘expressions’. Let
us call this shared property, for the purposes of the present discussion,
{EE}.
From what is known about linguistic diversity in present day speech
communities, we may safely conclude that, at any point of time before the
sixteenth century, the set of linguistically competent minds to be found
(10)23
he 1- he (6 7 h’), 2–3 hi, 2 heo, 3–4 e, ghe, 3 hæ, 3–4 ha, 4 ho, 3 e,
3–9 (dial.) a, 4–5 hye, 6 hie, 4–7 hee
23 The numbers before the cited forms represent the centuries of attestation, with ‘1’ indi-
cating the eleventh century, ‘2’ the twelfth, ‘3’ the thirteenth, and so on. Additionally,
‘1’ is exceptional in that it refers also to the centuries preceding the eleventh, from
which manuscripts in English have survived. A hyphen following a number identifies the
spelling that is still used in Modern English Standard.
Approaching ‘language change’ 45
for now to observe that after it had occurred, the rest of the ‘change’
must have been essentially a quantitative matter, that is, a change in
the distribution of competence properties within a population of minds.
It seems to me that this quantitative aspect is far more significant for
understanding linguistic evolution than the question how new properties
may arise within a population of competences in the first place. After all, a
competence property which never gets implemented in a sufficiently large
number of individual minds, will be as good as unnoticeable for speakers
and linguistic historians alike, and might as well not exist at all for that
matter. Therefore, the central question seems to be how a change in the
relative frequencies of properties within a population of competences may
come about, and what may cause it.
This question, it seems to me, breaks into two, more fundamental prob-
lems, namely (a) how do competences assume the properties they have,
and (b) how do they maintain them? Speaking very generally, problem
(a) can be given the following answer: competences get their properties
partly from their biological basis – which is genetically determined and
which imposes universal constraints on the properties that competences
may assume,24 and partly from the dynamical and complex interaction
between minds and environmental factors, which results in ‘language
learning’ or ‘language acquisition’. Question (b) can be answered along
similar lines. In order to be maintained, the properties of competences
depend both on their genetically provided biological substrate (such as
a working long term memory in a living body) and on interaction with
their external environment (linguistic competence that is not put to any
use whatever, for example, is unlikely to remain stable).
Now, among the factors in the environment of a linguistic compe-
tence that are likely to affect its properties, the experience of meaningful
discourse will certainly be more prominent than most others. In order
to experience meaningful discourse, however, a person depends on the
presence of people with whom they can interact via behaviour and text
production. In order for such interaction to be possible, of course, those
people need to be linguistically competent as well, and the discourse
they produce, understand and react positively to will clearly reflect, or
express, the properties of their own competences. Thus, the properties
which individual competences can assume and maintain will to a con-
siderable extent be determined – albeit indirectly – by the properties of
the competences in their environment, that is, in the population in which
they develop.
Of course, this is exactly the causal link chain which has been pro-
posed in section 3.3.5 above, and which is implicit to the view that lan-
guage(stage)s derive historically from earlier ones, and have a bearing on
the evolution of later ones. In short, competences are likely – and indeed
known – to assume properties that are similar, or – on a sufficiently coarse
graining – ‘identical’, to the properties of the competences their ‘owners’
interact with. One can therefore also say that (a) the competence proper-
ties to be found within a population get copied or replicated when they are
assumed by other (typically) new competences, and that (b) it is through
such replication that their stability within a population is brought about.
Thus, the question how a change in the distributions of {EE} and {II}
in representations of he, me, see, meet, green and so on was brought about
can be answered, very generally, by saying that {II}s were copied and
maintained both more effectively than, and at the cost of, {EE}s. Gener-
alising from the particular case, it can then be said that the question how
languages evolve and change boils down to the question how competence
properties emerge, how they replicate, and how, by doing so, they acquire
historical stability.
long-lived. Some will disappear while others will become more frequent,
so that it is conceivable (and probably even inevitable) that over time also
people’s understanding of what Beethoven’s Ninth is actually like will
change. Contrary to the changes that Mount Everest can undergo, the
course which the evolution of Beethoven’s Ninth will run is not a matter
of simple decay. Instead, it will reflect a complex variety of circumstantial
factors and be very difficult to predict.
Crucially, the differences between replicating and non-replicating items
have consequences when it comes to describing and explaining their prop-
erties. The case of Mount Everest is rather straightforward. If we want to
understand it, we will subject the mountain to a empirical investigation,
measure it, analyse its composition, and so on. We can then understand
it as the result of the seismic and volcanic forces that built it, and the ero-
sive forces that have given it its present shape. In the case of Beethoven’s
Ninth, however, it is not even obvious what it actually is that we should
describe: should we look at Beethoven’s autograph? Should we look at
contemporary interpretations? Should we look at all (or at least many) of
them and describe what they have in common? Obviously, we shall have
to make some decisions, however, and whatever they may be, it is clear
that we shall never be studying ‘Beethoven’s Ninth as such’ but always
particular instantiations of it.
If we study individual instantiations or copies, however, many of their
properties will be due to the fact that they are copies and thus to the
mechanisms and the circumstances of the processes by which they came
to be copied. Denying this would be indefensibly essentialist: only if
Beethoven’s Ninth ‘is his Ninth is his Ninth’, will differences between
individual instantiations be irrelevant and one copy ‘as good as another’.
But Beethoven’s Ninth is not simply ‘his Ninth’. There are only individ-
ual instantiations out there, and that we happen to regard them as tokens
of a single type is just our normal common sense way of dealing with
them. As soon as one acknowledges that only individuals exist, of course,
both the properties they share as well as the differences between them
do matter, and many of them are likely to reflect factors relating to their
replication.
Let me illustrate these claims through an example, and look at
the following reproduction of Andy Warhol’s Marilyn Monroe26 (see
figure 3.4 on page 51). Consider the peculiarities of your individual copy
first. For example, the fact that it is in shades of grey rather than in
colours follows from the design of the laser printer with which the copy
was produced. The amount of detail in the copy reflects the resolution
something about Andy Warhol’s Marilyn Monroe (both the original and
most of its copies) which has made it extremely successful as a replicator.
If this is so, it must – at least to some extent – be that fact which has
‘caused’ the reproduction above. It would then clearly be plausible to say
that your Marilyn is there because the particular combination of properties
that constitute it are a highly successful team of replicators. Thus, both
the existence as well as many of the properties of replicated objects can
be derived from the fact that they have been reproduced and from the
circumstances of their replication.
Similar arguments are valid for everything that people make copies of.
The quality of any particular rendering of Beethoven’s Ninth will both
reflect Beethoven’s original composition and its quality as a (team of)
replicator(s). Also, it will reflect the mechanics by which it is reproduced,
such as the instruments on which it is performed, the virtuosity of the
performers, or their ability to read musical notation. Likewise, the design
of an individual chair will reflect the way in which it was manufactured,
the purpose for which it was made, the creativity and skill of the artisan,
his or the prospective buyers’ aesthetic preferences, as well as their knowl-
edge of and the experiences they made with prior instances of chairs. As
far as apples are concerned, finally, the qualities of individuals will, on
the one hand, reflect the skills of farmers, the tastes of consumers, the
availability of fertilisers and pesticides and so on, and on the other hand,
the properties of those apples from whose seeds the ‘apple copies’ were
grown.
Obviously, the number of examples could be increased ad infinitum.
They all show that objects which are ‘reproduced’ by humans do owe
many of their qualities to the ways in which and the reasons why peo-
ple reproduce them. The properties of each individual copy appear to
be co-determined by the qualities of the ‘models’ after which they were
made on the one hand, as well as by inherent constraints and external
factors affecting the processes of their re-production. Thus, the relation-
ship between the properties of replicated items and the circumstances of
their reproduction is non-random and therefore informative. For exam-
ple, by looking at the above copy of Andy Warhol’s Marilyn Monroe, you
may be able to tell that it was copied to paper from a computer via a
laser printer. Conversely, you may be able to predict what the copy of a
picture will look like if you scan it, store it in JPG-format and then print
it out. This is because only some properties of any original are preserved
in a specific reproduction. Others will be lost, as in the case of Marilyn
Monroe colours and everything between the individual ‘pixels’ of which
computerised images are made up. On the other hand, if I were to copy
the digitised version that is stored on my hard disk, the resulting copy
Approaching ‘language change’ 53
would be, for all practical purposes, identical to its master. No further
information would be lost. This is important, because it highlights a cru-
cial difference between Andy Warhol’s original and copies in JPG-format,
namely that, when computers are the most commonly available means of
picture reproduction, the number of copies that look more like the one
on my screen will soon surpass the number of those that look more like
Warhol’s original. If from this day onward only digital copying were avail-
able, we could even predict that sooner or later only copies that look like
the one on my screen will be around. Since nothing lasts forever, all non-
digital ‘Marilyn Monroes’ will eventually decompose or be destroyed,
and all information that has not made it into the JPG-files will be lost.
From the quality of the pictures that they excavate future archaeologists
will then be able to reconstruct that there must have been a period on
our planet when all copying was done by computers, and they may even
infer the properties of those computers from the pictures they find. In
short, when one confronts objects or systems that exist through being
reproduced one cannot neglect that fact if one wants to understand why
they are as they are.
Clearly, languages are systems of this kind .
that the state of any language system at any time reflected essentially arbi-
trary social conventions, and could be understood better if one neglected
that it also had a history. While Saussurean structuralism, with its focus
on language states rather than language histories, certainly diverted lin-
guists’ attention from historical questions altogether, I do not think it
would have managed to bring a research programme more or less to a
halt, had there not been other factors involved. Instead, the main rea-
son why the linguistic community rejected the notion that the properties
of languages might be understood through the ways in which they are
replicated was that an approach which hinged on the concept of prop-
erty replication turned out to be immensely successful in another scien-
tific discipline, and eventually transformed it radically. That discipline
was biology, of course, and the approach was Darwinian Evolutionary
Theory. Arguably, the progress which biology has made since Charles
Darwin first described species of organisms as systems whose constituents
and properties owed their existence to the fact that they had been repli-
cated more successfully than competing variants has come to change our
view of the world and our role in it more radically than any other scien-
tific discovery before or after. Yet, it has not convinced linguists that they
should adopt a similar approach. Instead its success in biology may have
caused linguists to discard the whole idea more or less completely.
That this should have happened is due, I think, basically to two fac-
tors. First, as both Schleicher’s and Paul’s proposals go to show, the
linguists who first approached languages as replicating systems were not
cautious enough. Instead of slowly and subtly adapting the general idea
of property replication to the realm of language, they adopted concepts
and terms developed by biologists much too straightforwardly. Those
concepts were made to deal with biological phenomena, however, not
linguistic ones, so much unnecessary confusion was created through
rash analogical transfer. August Schleicher, for example, compared lan-
guages to biological ‘organisms’ and proposed that they went through
the developmental stages of infancy, youth, adulthood, senility and even-
tually death. Otto Jespersen (1922), on the other hand, compared them
to ‘species’, and assumed they were optimised and made continuously
‘fitter’ through their evolution.
Of course, it is obvious from our contemporary perspective that both
views are untenable for a variety of reasons (see the discussion in
McMahon 1994: 314–24). However, it is not their inadequacy that mat-
ters here but the reasons behind their inadequacy. To a large extent,
it seems to me, they are to be found in the scholarly impatience with
which conceptual frameworks for the study of language replication and
evolution were developed. Thus, both Schleicher and Jespersen seem to
Approaching ‘language change’ 57
Conference’, where not only was the question how the human language
faculty might have (biologically) evolved addressed, but also the question
if the histories of human languages should be conceived of in evolutionary
terms. And in 1996, to give a final example, the refereed online Journal of
Memetics33 was established on the internet, which deals with ‘evolution-
ary models of information transmission’. Clearly, human language falls
within its scope.
Of course, revived interest in an approach does not yet say anything
about its qualities, and the fact that other scholars are adopting perspec-
tives similar to the one advocated here does not automatically lend it any
plausibility.34 The question is rather why they are doing so and what they
expect to get out of it.
62
The Darwinian approach 63
they had been designed and created by a supernatural artisan. Hardly any
culture lacks a creation myth which includes such a deity, and until way
into the nineteenth century the complexity and functionality of almost all
things natural was adduced as a serious and seemingly irrefutable argu-
ment for the existence of a divine creator.3 Now, evolutionary biology
provides a theory which explains the same complexity and functionality
without having to make reference to supernatural agents which represent
huge explananda themselves. According to that theory (first put forward
in Darwin 1859), the amazing properties of living organisms result mostly
from two types of natural, albeit inherently complex, processes.
5 The sequence in which I’m presenting the two sub-theories actually reverses the order
in which they were discovered. Historically, the explanation why the inheritable elements
that underlie the properties of organisms had to be as they were was discovered before
one knew what genes exactly were, and how they worked in building cells and organisms.
6 This is true for most animal and plant species.
The Darwinian approach 65
this is not a biological essay, it would be beside the point to tell the
story in all its domain-specific detail. Instead, and since my account will
necessarily be schematic, I shall attempt to get it across in terms of a
thought experiment.
From this most of the rest follows. In order to see how, imagine a
bowl of free floating A, C, G and T building blocks in which you
throw a single strand of replicating DNA. From what has been said so
far, it seems to follow that this string will replicate until all the available
building blocks are used up. Eventually, you should end up with a bowl
full of identical copies of the original DNA strand, with a small quantity
of residual building blocks still floating around.
just as stable, our bowl will in the end be filled with a stable mix of
A–Ts, C–Cs and building blocks, the proportion of A–Ts against
C–Cs depending on when in the ‘history’ of the bowl the ‘copying
mistake’ took place and on how many free floating building blocks were
available at that time.
68 Selfish Sounds and Linguistic Evolution
9 Either because they are so long-lived that they can afford to take their time hunting for
building blocks, or because they are they are so quick at replicating that, in spite of the
A–T competition, more new copies are made than disintegrate.
The Darwinian approach 69
and (b) the fact that strings which are more stable and replicate better
than the rest necessarily oust the others, so that only those are ‘selected’
to remain in the bowl.
This is the essence of the Darwinian theory of evolution: when ‘random
mutation’ (that is, copying mistakes) and ‘natural selection’ (that is, what
comes about when some replicating molecules are better at replicating
or remaining stable than others) occur in a population of replicators,
the overall ‘fitness’ of that population will increase automatically and
without anybody steering the process or consciously trying to design fitter
variants.
synthesising a device which has the effect of decomposing Bs into raw
material for replication. Apart from that, Bs are better at replicating
than As because they chemically attract building blocks. Cs, finally,
repel As, so that the latter cannot ‘get at them’, but happen to repel
many building block types as well. Thus, while being relatively stable,
Cs are slow at replicating. To make the scenario easier to imagine, a
graphic representation is given in figure 4.1.
We can now predict that if there were only As and Bs around, the
former would eventually out-replicate the latter and oust them from the
bowl. Equally, we can predict that if there were only Cs and As around,
As would oust Cs because even though they cannot come close enough
to decompose them, they do not repel any building blocks, will replicate
more quickly than Cs and eventually outnumber and oust them. A sim-
ilar relation will hold between Bs and Cs. Bs do not threaten Cs
directly, but their capacity of attracting building blocks will eventually
deprive Cs of raw material for replication. Thus, Cs could not suc-
cessfully compete against either As or Bs alone.
The Darwinian approach 73
Comment: The CB alliance is stable in the presence of As. Cs ‘pro-
tect’ Bs from attacks by As, and Bs ‘help’ Cs by attracting building
blocks. As soon as the number of As sinks below a critical point, the
CB alliance will break, Bs will out-replicate Cs, and, ‘unprotected’
will provide easy victims for the As that are still around, so that the
latter will multiply again, forcing Bs back into alliances with Cs.
Thus, the proportion of As, Bs and CBs will keep fluctuating
around a stable equilibrium.
Figure 4.2 The evolutionarily stable distribution of As, Bs and Cs.
But if the three of them happen to be around at the same time, this
changes the matter completely. In the vicinity of Cs, Bs will thrive,
because Cs keep As at a distance, so that they cannot get at Bs
anymore. At the same time, Cs will profit from the ability of Bs to
attract building-block material, but again only as long as Bs and Cs
are close enough to each other. One possible outcome of such a scenario
would then be a stable equilibrium in which all of the three replicator
types co-exist. Bs will only replicate successfully as long as there are
Cs around, which repel ‘B-eating’ As. Cs, on the other hand, can
only replicate successfully if there are Bs around, which attract enough
building blocks so that Cs are not out-replicated by As. And finally,
also As will manage to replicate successfully in this scenario, because
as soon as Bs start to out-replicate Cs, there will be a sufficiently
large number of ‘unprotected’ Bs for As to decompose. We may thus
predict the stable replicator population in the region to be composed of
CB complexes, free floating Bs and free floating As, as illustrated in
figure 4.2.
74 Selfish Sounds and Linguistic Evolution
and the genes ‘sharing’ a position may differ. Sometimes only one of the
two candidates (the so-called ‘dominant’ gene) gets expressed, while the
expression of the other (called ‘recessive’) is suppressed. At other times,
the phenotype expresses both genes, which yields a mixed phenotypic
property. But all this concerns primarily development. What matters for
evolution is that half of the genes in the total genome of an organism are
inherited from the organism’s mother, and the other half of them from
its father. These halves are provided in the form of paternal and maternal
sex cells ( ‘gametes’), which have to get together when a new organism is
to be formed. Crucially, when parental genomes divide in halves to form
gametes, each gamete gets an idiosyncratic combination of genes from
the two teams in the parental genome. Since the number of genes in a
genome is high, and the number of possible combinations even higher,
no two organisms are likely to have identical genomes (apart, of course,
from identical twins).
Now, since normally only gametes of co-speciates can combine to form
a new organism, it is possible to define a species as comprising only those
individuals whose genomes are combinable in this way. Although not
unproblematic either (see the discussion in Eldredge 1995: 106–20), this
definition appears to be less fuzzy than one based on the number of shared
genes, or (worse) phenotypic similarities, for instance. Thus, species seem
to represent units which do not seem to be derivable from the replicator
level.
However, this may be only superficially true, and the strange combina-
tion of internal variability and relative integrity that characterises species
may after all make sense from the point of view of the replicators involved.
Could there be a reason why an individual replicator might be able to
replicate better or to exist longer, if the composition of the teams with
which it lives and replicates does fluctuate? Imagine a genome which has
reproduced faithfully and a-sexually from its parent. Every gene on that
genome can be sure, metaphorically speaking, that the team in which it
exists represents a viable combination, otherwise it would not have been
replicated and would not exist. At the same time, teams of different com-
positions might be just as viable, or even more so, and the gene would
replicate better, if it was in them. From an individual gene’s point-of-
view, it would therefore make sense to try out new combinations. After
all, it has no loyalty to any of its present team mates. On the other hand,
new combinations may also be unviable, and since there are more ways
of being dead than of being alive, changing teams appears too much of a
risk to be worth undertaking after all. Thus, sexual reproduction should
not really occur.
What, however, if a gene was able to tell which potential new partners
are likely to be safe and promising? Clearly, in this case, the risks of
The Darwinian approach 77
13 Incidentally, the fact that in sexual reproduction genomes do not copy faithfully explains
why genes, and not they, are the primary players in evolution. The high degree of vari-
ability of individual genomes, makes it impossible to say that some type of genome is
more stable, or produces more copies than others. There simply are no genome ‘types’.
Each of them is slightly different from all the others, and even if it is very successful in
terms of the life-span of the organism it codes for, it will not usually find its way into the
next generation intact.
78 Selfish Sounds and Linguistic Evolution
14 Evolutionary theory represents not only an epistemological bridge between two sciences,
however. The link between chemistry and biology it establishes seems to be describing
an actual historical development as well. After all, it tells the story about how life may
have emerged from chemistry under the special circumstances on our planet, and it does
so without having to make reference to metaphysical forces such as divine creators or
mysterious substances such as elan vital.
15 The analysis which would produce the raw data for such a computation would involve
immensely fine-grained measurements, and the resulting number of factors that would
then have to be fed into the calculation would be so great and the interaction-laws so
complex, that we will face one of those proverbial situations in which all the computers
on earth would take longer than the actual age of the universe in order to finish the job.
82 Selfish Sounds and Linguistic Evolution
Thus, while the boundaries of bodies might be ‘fuzzy’ from the genes’
point-of-view, they are still boundaries, and genes are more likely to
respect them than not.16 In so far as the evolutionary fate of genes depends
on whether their bodies manage to reproduce before disintegrating, it is
justified to say the relationship between bodies and genes works both
ways, even though much more causal traffic may flow from the bottom
up than from the top down.
Similarly, although sexual reproduction and species may have emerged
for the benefit of genes, their existence has in turn come to impose con-
straints on the evolution of genes, so that sometimes the species-level may
have to be invoked for describing and explaining it. The following argu-
ment (based on Eldredge 1995: 106–39 and following Paterson 1985)
may serve to illustrate the point. Organisms normally seek their mates
for reproduction among co-speciates. Their ability to recognise them is
governed by what may be called their mate recognition system. Now,
imagine a mutant organism, say a dog, which sports a third eye on the
back of his head and which has therefore 360-degree vision. Naively, we
might expect this dog to live longer and to produce more offspring than
its co-speciates so that in the long run only three-eyed dogs will be found
in the population. However, because of its third eye the dog might no
longer be recognised as a dog by the females in the population. They may
avoid mating with him, and seek other partners. Thus, for all the advan-
tages of having 360-degree vision, the dog will not reproduce and the
gene for the third eye will die out with him. Therefore, the fact that genes
have organised themselves into the separate pools which we call species,
has introduced a constraint on their evolution which clearly deserves to
be called top-down, even though it does not warrant the conclusion that
evolution works for the benefit of species.
complex ways, from chemical ones. Also, the theory itself shows why some
aspects in the historical unfolding of life will have been truly accidental,
or at least as close to accidental as makes no difference. Since innovations
come about through ‘copying mistakes’, evolution proceeds blindly, that
is, by trial and error, merely preserving ‘adaptive’ mutations once they
have occurred and discarding those that turn out not to be viable. Thus,
evolution has no foresight, and cannot be expected to ‘know’ which muta-
tion would be optimally adaptive under which circumstances. Therefore,
it might just as well not provide it. Whether or not it actually does, appears
at least to some extent to be a matter of chance, precluding the possibility
of a non-probabilistic, predictive theory by definition.
to acquire evolutionary stability have not done so because they are opti-
mal, but simply because they are viable, and fitter than the competitors
which accident has pitched them against. This implies that local trends
that conspire towards the optimisation of particular functional proper-
ties, such as camouflage, speed, or sight are definitely likely to occur, but
generally speaking, evolutionary developments have never conspired to
produce what might be considered ‘best’ in absolute terms. Instead, they
have selected what has turned out to be good enough for survival and
discarded what is not. In order to make this clear, evolutionary biologists
prefer to speak of ‘satisficing’ rather than optimal properties.
Furthermore, evolution is by its very definition constrained to work
on the basis of what is already there. Its agents are copies, and copies
are always modelled after something that has been there before them.
Thus, evolution never designs genomes from fresh, but creates novelty
by adding to or altering given designs. As it has been aptly put by Konrad
Lorenz (1973: 25), the resulting systems, impressive though they are,
must also always to some extent be bricolage, like houses which are repeat-
edly remodelled over time, and which are full of items that may have been
of better use at earlier times, or which have simply been retained because
they have not proved to be downright harmful.18
the genetic one. Sometimes, this environmental change may itself have
an evolutionary explanation, as when predators and prey enter into evo-
lutionary arms-races and evolutionary changes in the former trigger evo-
lutionary responses in the latter, or vice versa. Sometimes, however, the
causes of an environmental change may lie outside the field of biology
altogether, as in the case of the meteor which – allegedly – caused the
extinction of the dinosaurs and paved the way for the mammalian area.
Now, it is obvious that biologists should be more interested in such
changes where environmental conditions are – for practical purposes –
more or less constant and most of the relevant action takes place within
the gene-pool of either a single species or a select set of species within
a defined ecosystem. After all environmental changes may have non-
biological causes. Therefore, catastrophic ‘external’ events like meteor
impacts, or climatic changes are often regarded as mere historical acci-
dents, contingencies which do themselves not call for explanations, at
least not for biologically interesting ones. Also, it seems to be the gen-
eral opinion that environmental catastrophes have been relatively rare
and separated by long periods in which populations of replicator teams
(i.e. genomes) were left to play it out among themselves, so to speak,
and to become gradually better and better adapted to conditions which –
from their point of view – did not change much. Explanations of the
complex properties of biological species typically highlight the fact that
they have resulted from a long series of small evolutionary steps in which
successive generations of better adapted replicator teams ousted their less
fit predecessors, while the potential impacts of environmental conditions
are typically relegated to the notorious ceteris paribus condition. Thus,
Richard Dawkins’ provocative answer to the question ‘Why are people?’
is ‘because genes for making people have managed to replicate before
disintegrating’ (1989: chapter 1), and not ‘because a meteor hit the earth
some 50 million years ago’. Clearly, the latter answer would be as justified
as the first, the crucial difference being not its plausibility but the fact that
it does not exactly invite further research. It boils down, in a way, to say-
ing that we are essentially here because of a coincidence, which may be
true but fails to stimulate further enquiry. It is important to remember,
though, that Darwinian accounts of evolution which focus on competi-
tions among rivalling replicator teams within populations and/or sets of
such tend to give us only half of the story, even if it happens to be the
more interesting one.
5 Generalising Darwinism
To the extent that languages in time share certain properties with other
classes of systems simply by virtue of being historical, there is no need
to invoke any ‘special’ local properties in order to characterize their
behaviour.
(Roger Lass 1997: 390)
89
90 Selfish Sounds and Linguistic Evolution
r Just as only co-speciates can have offspring with one another, only
speakers of ‘the same language’ can communicate.
r Just as living organisms can be described on the genotypic and on the
phenotypic levels, languages can be described on the level of speakers’
competences and on the level of texts. Texts seem to express compe-
tences just as phenotypes express genotypes.
r By this rationale, verbal behaviour, or performance, would be the lin-
guistic counterpart to embryological development.
r Like development, performance is always strongly co-determined by
external, or environmental contingencies. It is as difficult to predict,
from the properties of a person’s linguistic competence, what that per-
son will say on any particular occasion, as it is to foresee what an adult
organism will be like, if one knows only the properties of its genome.
r As far as the reproduction of both organisms and languages is con-
cerned, neither of the two can bypass the phenotypic level. Just as
genomes need to make themselves organisms to reach the next gen-
eration, so languages depend on being spoken if they want to make it
into the brains of new speakers.
Analogies like these are abundant (see for instance Stevick 1963, or
Lass forthcoming) but can only be carried to a certain point. Sooner
or later one will inevitably reach the limits of analogical transfer. Likely
mismatches are easy to think of:
r DNA replication seems to be a rather mechanical process, in which indi-
vidual sequences are copied not only faithfully but also very directly.
The replication of competences must be much more indirect and multi-
faceted. No immediate links between the competences of any two speak-
ers are conceivable, since brains have no direct access to one another.
r Similarly, straightforward competence lineages do not seem to exist
either. While in asexually reproducing species every organism has
exactly one parent, and in sexually reproducing species not more than
two, speakers normally acquire their languages from a large number
of individuals. The term ‘mother tongue’ is absolutely misleading in
this regard, and it is impossible, for all practical purposes, to determine
which individuals should count as a speaker’s linguistic ancestors.
r Furthermore, the genomes of organisms do not change after concep-
tion. Basically, organisms keep their genomes during their whole lives.
A person’s language, by comparison, does not ever seem to stop chang-
ing at all. Although our competences do assume a certain stability once
we reach the age of ten, roughly speaking, we can learn new words,
idioms, grammatical constructions and pronunciations also after that.
r Another fundamental difference between biological and linguistic evo-
lution is that the former is blind to the characteristics an organism
Generalising Darwinism 91
1 For essays and studies on that topic see for instance Boyd/Richerson (1985), Cavalli-
Sforza/Feldman (1973, 1981), Cloak (1973, 1975), Dunbar/Knight/Power (eds.) (1999),
Gould (1982), Holland (1975), Hull (1988), Hurford/Kirby (1997), Lumsden/Wilson
(1981), Maynard-Smith (1989, 1996), McPeek et al. (eds.) (forthcoming), Plotkin (ed.)
(1982).
92 Selfish Sounds and Linguistic Evolution
biological evolution is just one of many instances of cumulative blind variation and
selection leading to the adaptation of one system to another. So although scientific
theories, cultural practices, and genes may exist in very different forms and employ
distinct modes of variation, selection, and replication, these and other superficial
differences have in themselves little bearing on the argument that both thought
and science make progress through a process of cumulative blind variation and
hindsighted selection. [. . .] Biological evolution, insofar as it leads to increases
in adapted complexity, is a selectionist process. But not all selectionist processes
have to mimic adaptive organic evolution in all of its biological details. (Cziko
2000: 287)
1. prebiotic chemical evolution, including the chemical processes that gave rise
to terrestrial life around four billion years ago, others that must have given
rise to various life-like assemblages elsewhere in the universe, and related
chemical processes that can be studied in the laboratory, with the potential
for developing products of great utility;
2. biological evolution, leading through mutation and selection to the enormous
variety of life forms on earth and also to the existence and evolution of ecosys-
tems;
Generalising Darwinism 93
2 Note the frequency with which terms like evolve, evolution, mutation, and selection occur in
all but one paragraph of Murray Gell-Mann’s overview. This indicates the paradigmatic
character which biological evolution has assumed within the interdisciplinary research
programme, but should not be taken to imply that all other domains are seen, metaphor-
ically, through biological spectacles.
94 Selfish Sounds and Linguistic Evolution
CONSEQUENCES
(AT ‘PHENOTYPIC’ LEVEL)
SELECTIVE
EFFECT
ON
VIABILITY
OF SCHEMA
AND
ACTUAL PREDICTION, BEHAVIOUR
COMPETITION
(AT ‘PHENOTYPIC’ LEVEL) AMONG
SCHEMATA
COMPRESSION
PREVIOUS DATA,
INCLUDING PREVIOUS BEHAVIOUR AND ITS
EFFECTS
[. . .]
In the case of a scientific theory, the feedback comes from the comparison of
the results of calculation with experiment. In biological evolution, the success in
producing offspring of an organism with particular DNA (or the relatives of that
organism) affects the survival of that or a similar pattern of DNA.
For individual learning and thinking, schemata can be ideas, including creative
ideas, or patterns of thought that are ways of interpreting the world. The results
of the behavior generated by those ideas or patterns in particular situations can
influence how those ideas or patterns fare in competition with others. (1992: 10f.)
Since our focus is on biological and linguistic systems, let us try to find
out in some more detail, albeit tentatively at this stage, how Gell-Mann’s
schematic diagram can be applied to life and to languages. All we have to
do, it appears, is to specify the elements that fill the individual slots.
CONSEQUENCES
(COSTS, DEATH, REPRODUCTION, ETC. )
SELECTIVE
EFFECT
ON
VIABILITY
OF SCHEMA
PHENOTYPIC EXPRESSIONS AND
(BODY + BEHAVIOUR) COMPETITION
AMONG
SCHEMATA
COMPRESSION
PREVIOUS DATA,
INCLUDING PREVIOUSLY EXPRESSED
PHENOTYPES AND THEIR FATES
they live, but also in terms of the number of instantiations that exist within
a population, or species as a whole, since the environment may ‘grant’ an
organism anything from zero to a large number of offspring.
Of course, the two kinds of feedback are related and their effects cannot
be neatly disentangled from each other, as the number of offspring which
an organism may have will often correlate with the time for which it is
alive, for example. But in extreme cases the difference does matter. A
genome might code for an extremely long-lived, but infertile organism,
for example. In that case the environmental feedback on all its somatic
instantiations would be positive, while the feedback on the total number
of instantiations with the population or species would clearly be negative.
On an evolutionary time scale, the genome type will be extremely short-
lived, no matter for how long the actual organism for which it codes may
survive.
Now, since the environmental feedback which intra-somatic genome
instantiations receive has no variety to select from (the genome copies in
body cells are, for all practical purposes, of the same type), no ‘evolution’
or ‘learning’ can take place on the level of an individual body’s genome.4
Instead, the systems that ‘evolve’ or ‘learn’ in the case of biological evo-
lution will be populations of genes, and it is the environmental feedback
on such populations rather than that on the genes of individuals which
drives this. What adapts to the environment is the distribution of genome
types within the gene-pool that defines biological species, and it is on this
level that biological evolution amounts to a learning process.
4 This does not mean the development of an individual body does not involve ‘learning’,
‘adaptation’, or ‘evolution’ in other respects. But in such cases, the ‘schemata’ among
which environmental feedback can come to select will not be genetic. Instead, they will
involve populations of antibodies, as in the case of immune system evolution (see Clark
1995), or patterns of synaptic connections as in the development of the central nervous
system and cognition (see Edelman 1987, Plotkin 1994).
Generalising Darwinism 99
CONSEQUENCES
(COSTS, SUCCESS, ETC.)
SELECTIVE
EFFECT
ON
VIABILITY
OF SCHEMA
AND
VERBAL BEHAVIOUR COMPETITION
(PERFORMANCE) AMONG
SCHEMATA
COMPRESSION
PREVIOUS DATA,
INCLUDING PREVIOUS COMMUNICATIVE
BEHAVIOUR AND ITS EFFECTS
may, or may not ‘match’ and so on. Evidently, minds will have ways
of assessing the benefits gained by individual speech acts as weighted
against the costs incurred by them, for instance in terms of processing
and/or articulatory energy.
Language acquisition can thus be regarded as an evolutionary pro-
cess in which a specific part of a person’s mind/brain (Chomsky’s UG,
Pinker’s ‘language organ’) ‘adapts’ to aspects of its host’s body, to the
conceptual content of other parts of that host’s mind/brain, as well as
to the external environment of its host, in particular to the communica-
tive behaviour of the social group within which its host is embedded.
Consequently, a person’s linguistic competence evolves into a schematic
representation of these aspects. Figure 5.3 above, schematises that
interpretation.
Generalising Darwinism 101
, ɔ mon
5 See, for example, Lass (1980, 1987b, 1997), Dressler (1985), Samuels (1987a, 1987b).
102 Selfish Sounds and Linguistic Evolution
6 The issue has been hotly contended within the linguistic community. Interesting con-
tributions are, for example, Lass (1980, 1987a, 1987b, 1997), Dressler (1985: 265–8),
Samuels (1987a and 1987b).
Generalising Darwinism 103
CONSEQUENCES CONSEQUENCES
(COSTS, SUCCESS, ETC.) (COSTS, SUCCESS, ETC.)
SELECTIVE SELECTIVE
EFFECT EFFECT
ON ON
VIABILITY VIABILITY
OF SCHEMA OF SCHEMA
AND AND
COMPETITION VERBAL BEHAVIOUR VERBAL BEHAVIOUR COMPETITION
AMONG (PERFORMANCE) (PERFORMANCE) AMONG
SCHEMATA SCHEMATA
COMPETENCE STATE THAT SUMMARISES THE CONDITIONS UNDER COMPETENCE STATE THAT SUMMARISES THE CONDITIONS UNDER
WHICH PAST COMMUNICATIVE BEHAVIOUR AND PREDICTS TYPES OF WHICH PAST COMMUNICATIVE BEHAVIOUR AND PREDICTS TYPES OF
SUCH BEHAVIOUR ARE SUCCESSFUL SUCH BEHAVIOUR ARE SUCCESSFUL
(ONE OF MANY COMPETENCE STATES RELATED BY MUTATION AND (ONE OF MANY COMPETENCE STATES RELATED BY MUTATION AND
COMPETITION) COMPETITION)
COMPRESSION COMPRESSION
Legend: : Pathway for feedback from the competence of speaker A to the competence population.
: Pathway for feedback from the competence of speaker B to the competence population.
Figure 5.4 Language evolution and change, viewed as a Complex Adaptive System.
Generalising Darwinism 107
A’s mind. At the same time, the produced utterance will trigger a con-
comitant unfolding of speaker B’s current competence state, resulting in
‘interpretative’ performance. Once again, this interpretation will incur
costs in terms of processing energy, and information about this will feed
back to speaker B’s mind, where it will be available for assessment. Finally,
the utterance and its interpretation will have further effects on the subse-
quent behaviour and cognitive states of both involved speakers, affecting
their actions, social relations, the ways in which they assess the situation,
and so on. These consequences will also feed back to the involved speak-
ers’ minds/brains and the competence states implemented in them. In
each speaker’s mind, they will be ‘evaluated’ separately, weighted against
the costs incurred by articulation and interpretation. Consequently, they
will either stabilise, or destabilise the current competence states. If both
participants’ minds/brains evaluate the overall results of the communica-
tive exchange that are fed back to them positively, both competence states
will be reinforced. If either of them does not, the state of his/her compe-
tence will be destabilised and rivalling states be strengthened. Thereby,
individual competences exert ‘selection pressures’ on each other.
If such exchanges take place repeatedly, the parallel ‘acquisitional’ (and
as we have said above also ‘evolutionary’) processes that take place within
each of the speakers’ minds will select competence states that incur pos-
itive7 feedback when they unfold in communication. Presumably, the
competences in each of the involved speakers’ minds/brains will evolve
towards stable states in which they resemble each other fairly well.
Of course, the scenario just given is simplified in many respects and
raises a couple of questions that cannot be dealt with exhaustively at this
point. Some of them nevertheless deserve to be mentioned.
First, it is obvious that the amount of influence which two competences
exert on each other when they interact via their unfolding will hardly ever
be distributed symmetrically. Obviously, for example, children’s compe-
tences are more likely to adapt to those of adults than vice versa.
Second, no speaker communicates with only one other speaker, but
with many. To the extent that there are differences among the compe-
tences of different speakers, every single one of them will be exposed
to conflicting selective pressures, and will evolve towards a state which
represents an acceptable compromise among them.
Third, what makes for an ‘acceptable compromise’ is a subtle question.
It is inconceivable, for instance, that a speaker’s competence state will
7 Of course, ‘positive’ must be understood in relative terms. What matters for the selection
of a particular competence state is that the feedback it gets is more positive than that of
rivalling states.
Generalising Darwinism 109
5.2.2.3 Summary
We have seen that languages can be construed as specific subtypes of a
more general class of systems, which are adaptive and thus capable of
110 Selfish Sounds and Linguistic Evolution
then also linguistic knowledge must emerge from the interactions of neu-
ronal configurations. Like biological replicators, then, these specific con-
figurations exist because their properties and effects stabilise them, and
cause them to replicate before they perish.
Of course, the conclusion that competences might not merely be pas-
sive objects which speakers acquire and use as tools for their communica-
tive and cognitive purposes strikes one as rather counterintuitive at first.
Generalising Darwinism 111
which are defined by social networks (see e.g. Milroy 1980). However, it
has the same implications as the view that speakers change their languages
because they falsely believe they can improve them. First, it begs the ques-
tion by what features they should decide to distinguish themselves from
their neighbours. Partly, this will be constrained by what these neighbours
happen to be doing of course, but this transfers the problem merely to
the next level. Also, the question still remains: which of the possibilities
to ‘improve’ the language that is not taken by their neighbours are they
supposed to take? If their decision is purely arbitrary, then this puts an
end to all attempts at understanding language change; if it is not, but
instead motivated by properties of the language they happen to speak,
then speakers lose their autonomy to exactly that extent.
What this means, however, is that there can be no functionalist account
of linguistic evolution in which speakers are viewed as autonomous and
free-willed agents of change and languages merely as passive objects.
There is no way of avoiding this conclusion, and the fact that it becomes
so obvious if one views languages as complex adaptive systems may be
one of the greatest advantages of that perspective.8
8 Interestingly, the view that speakers might not be irreducible agents of linguistic change is
implicit in many functionalist accounts of language change anyway. To give an example,
explanations of phonological lenitions are hardly ever explicitly attributed to a decision on
behalf of speakers to make pronunciation easier for themselves. Instead explanations are
given in terms of depersonalised concepts such as ‘inertia’, which is ultimately a physical
principle over which speakers can definitely have no control. The same holds true, of
course, for the relative ease with which different types of sounds are perceived. Again, the
perceptibility of acoustic signals is clearly not under speakers’ control, but represents a
genetically determined part of their physiological make-up. Similarly, even less obviously
mechanic aspects of language are typically couched in a-personal, often semiotic terms.
Thus, the iconicity or transparency of signs (such as complex word forms, that represent
complex conceptual configurations) is normally measured in inter-subjective terms. In
sum, one gets the impression that when regularities are expressed and explanations pro-
posed, no active role is attributed to speakers at all. The only occasion where ‘speakers’
are really adduced is when it comes to explaining, on a meta-theoretical level, why the
laws which are proposed to account for language properties, both synchronically and
diachronically are as weak as they typically are, that is, statistical rather than covering.
Human actions, it is then typically argued, will forever remain somewhat mysterious and
will never be fully understood.
One consequence of this attitude is that laws which yield wrong predictions are not
really to be regarded as problematic. Clearly, this may easily tempt one to be satisfied
with rather half-baked accounts. Indeed, hardly a conference goes by, for example, with-
out papers being given that contradict each other in differing degrees of blatancy. While
the obvious conclusion that both cannot be right is definitely drawn by many members
in the respective audiences, it tends to be quickly superseded by the melancholy and
normally unexpressed acknowledgement that probably both will be wrong, but as long
as we are having fun speculating about it, it does not really matter, because who will ever
really understand speakers? In sum, the assumed ultimate unaccountability of human
behaviour tends to lend itself as a good excuse to practise linguistics as the art of thinking
up elegant and aesthetically pleasing just-so stories.
Generalising Darwinism 115
At any event, if one gets accustomed to the idea that the properties
of languages might be described and to some extent understood without
deriving them from their supposed ‘owners’, that is, human selves, one
will see the parallels between biological and linguistic evolution much
more easily. Both in species and in languages, ‘evolution’ is effected
through environmental feedback on rivalling schemata, which ‘stabilises’
and/or ‘selects’ some at the cost of others. While the results of this selec-
tion process appear to be ‘goal-oriented’, ‘adaptive’, or ‘functional’, there
is nothing intentional, or teleological about them. Being patterns and thus
not identical with their material realisations, both biological and linguistic
schemata may remain in existence even as their actual material substrates
get replaced over time. This is what makes them ‘replicating’ systems. If
you define a Darwinian system as one which evolves by producing vari-
ety in a pool of replicating patterns among which some will prove more
stable and better at replicating than others under specific environmental
conditions, then languages are clearly just as ‘Darwinian’ as biological
species.
are made of DNA, however, but that they are replicators. Recall that we
have explained the essentials of Darwinian evolutionary theory without
the chemical details of DNA replication. As a matter of fact, some of
the thought experiments we conducted involved fictitious replicators that
bore no resemblance to DNA molecules at all. This very circumstance
might have made us suspect that Darwinian evolution was not dependent
on the existence of DNA.
This conclusion has also been drawn by Richard Dawkins, one of the
most prominent members of the contemporary biological community. In
his best-selling introduction to gene-based evolutionary theory he pro-
posed that the principles which made Darwinian evolution work would
do so wherever true ‘replicators’ emerged. ‘What, after all, is so special
about genes?’, he asks.
The answer is that they are replicators. The laws of physics are supposed to be
true all over the accessible universe. Are there any principles of biology that are
likely to have similar universal validity? When astronauts voyage to distant planets
and look for life, they can expect to find creatures too strange and unearthly for us
to imagine. But is there anything that must be true of all life, wherever it is found,
and whatever the basis of its chemistry? [. . .] Obviously I do not know but, if I
had to bet, I would put my money on one fundamental principle. This is the law
that all life evolves by the differential survival of replicating entities. [. . .] The
gene, the DNA molecule, happens to be the replicating entity that prevails on our
own planet. There may be others. If there are, provided certain other conditions
are met, they will almost inevitably tend to become the basis for an evolutionary
process. (1989: 191f.)
We have seen that competences evolve by the differential survival of
their constituent properties. If Dawkins is correct, then these should rep-
resent ‘replicators’. Dawkins himself suggests that they may. His argu-
ment continues like this.
Do we have to go to distant worlds to find other kinds of replicator and other,
consequent, kinds of evolution? I think that a new kind of replicator has recently
emerged on this very planet. It is staring us in the face. It is still in its infancy, still
drifting clumsily about in its primeval soup, but already it is achieving evolutionary
change at a rate that leaves the old gene panting far behind.
The new soup is the soup of human culture. We need a name for the new
replicator, a noun that conveys the idea of a unit of cultural transmission, or
a unit of imitation. ‘Mimeme’ comes from a suitable Greek root, but I want a
monosyllable that sounds a bit like ‘gene’. I hope my classicist friends will forgive
me if I abbreviate mimeme to meme. [. . .] It should be pronounced to rhyme
with the word ‘cream’.
Examples of memes are tunes, ideas, catch-phrases, clothes fashions, ways of
making pots or building arches. Just as genes propagate themselves in the gene
pool by leaping from body to body via sperms or eggs, so memes propagate
themselves in the meme pool by leaping from brain to brain via a process which,
in the broad sense, can be called imitation. (1989: 192)
Generalising Darwinism 117
11 As Matt Ridley pointed out, of course, the ‘Invisible Hand is exactly the same concept
as Natural Selection’ (2000: 28). The main difference is that the bottom-up approach
taken by Darwin makes it more easily possible to investigate the mechanisms by which
‘the Invisible Hand’ does its tricks, and thus to explain why particular ‘selections’ are
being made.
12 It is somewhat curious that neither Dawkins, nor any of the scholars who have taken his
idea seriously, has ever attempted to test its plausibility by applying it to the rich and well
documented data provided by linguistic change. After all, there are few other domains
of cultural change which have been studied in comparable detail and in comparably
systematic ways. Possibly, however, it is the very long tradition of historical language
studies as an academic discipline, and the progress they have made without employing
an evolutionary paradigm, which made outsiders like Dawkins, Dennett, Hofstadter,
Blackmore and many others cautious. After all, trying to contribute to a discipline with
whose theoretical assumptions and methodologies one is not very familiar, is always
risky. Linguists, on the other hand, might have felt that even though many problems
120 Selfish Sounds and Linguistic Evolution
The route towards understanding language change will then lead via
the following questions. First, we elaborate the notion of linguistic repli-
cators. We need to ask whether the competence properties that can be
inferred from linguistic behaviour and textual output qualify as such.
Secondly, we must try and determine in as much detail as possible how
competence properties actually do get replicated, and on what factors
their stability depends. Thirdly, we shall have to discuss to what environ-
mental feedback the stability and replication of competence properties
are likely to be sensitive. If we manage to give plausible – if only prelim-
inary – answers to each of the three questions, we shall have outlined a
theory of linguistic evolution or an evolutionary theory of language.
concerning the explanation of linguistic change were not satisfactorily resolved, they
were not so crucial as to necessitate the adoption of a paradigm which to many linguists
was as unfamiliar as the peculiarities of language change were to a biologist.
Generalising Darwinism 121
1 Importantly, this question does not depend on finding a way of physically describing
linguistic replicators. What matters is there may exist, in principle, mental linguistic
replicators capable of getting an evolutionary process going. Recall at this point also
that Darwinian evolutionary theory was established without the actual chemical units of
selection being known. Darwin himself did not grasp the importance of Mendel’s exper-
iments for his theory, and genes were discovered only in the twentieth century. And even
today defining the gene is not easy and in fact the term is used quite differently by breed-
ers, geneticists and molecular biologists because they are interested in different things. At
the molecular level, genes consist of sequences of nucleotides along a molecule of DNA.
Names are given to different lengths of DNA, such as a codon, which is a sequence of
three nucleotides, or a cistron, which is a sufficiently long sequence of nucleotides to pro-
vide instructions for building one protein – with a start symbol and a stop symbol. Neither
of these is necessarily passed on intact in sexual reproduction and neither corresponds
to what we think of as the gene ‘for’ something. [. . .] Yet it is these effects that natural
selection gets to work on. So what is the unit of the gene? [. . .] One useful suggestion is
that a gene is hereditary information that lasts long enough to be subject to the relevant
selection pressures. [. . .] This intrinsic uncertainty about just what to count as a gene
has not impeded progress in genetics and biology. It has not made people say, ‘We cannot
decide what the unit of the gene is so let’s abandon genetics, biology and evolution.’ These
sciences all work by using whatever unit they find most helpful for what they are doing at
the time. (Blackmore 1999: 54)
122
Towards an evolutionary theory 123
the glass pane of a copying machine is not a replicator. It may be annoyingly long-lived
(difficult to remove), and will copy fecundly (whenever the machine is operated) and as
faithfully as the resolution of the copier admits. Yet, nothing in the particular pattern of
the spot bears any causal relationship to the accidental fact that copies of it happen to be
produced. Therefore, although it is replicated, it is not active. Genes, on the other hand,
do play an active role in their own reproduction. That is to say, although their repro-
duction also depends on machinery, on chemicals in living cells, that machinery would
not reproduce anything but genes and it would not reproduce genes either, were it not
for their special properties. Thus, the fact that genes do have the properties they have is
essential for their replication. Therefore, even though their replication is something that
happens to them, it also happens through them, so that they truly deserve to be called
active replicators.
Towards an evolutionary theory 125
‘semantic features’ etc.; and (c) rules that relate these constituents in
various ways.5
Like meaning, code is seen to consist of hierarchically ordered con-
stituents, so that larger bits are constituted of smaller bits, and the ‘items’
or ‘building blocks’ of which code is made can be both complex and
simple. In a morphosyntactic hierarchy, for example, sentences consist
of phrases, phrases consist of words, words consist of morphemes, mor-
phemes consist of phonemes, and so on. The building of such hierarchies
may involve rules like S → NP VP, NP → (Det) N, VP → V NP etc.
Also, syntactic embedding allows complex constituents to assume the
roles of smaller bits in bigger structures, so that syntactic hierarchies dis-
play a certain degree of recursiveness, as in I hit the dog that chased the cat,
thought to be structured somewhat like this:
(14) [S [NP [N I]N ]NP [VP [V hit]VNP [NP [Det [the]Det [N dog]N ]NP
[S [NP [N which]N ]NP [VP [V chased]V [NP[Det the]Det [N cat]N ]NP ]VP ]S ]
NP ]VP ]S .
There are not only syntactic hierarchies, of course, but also others, such
as prosodic ones, in which utterances are seen to consist of intonation
phrases, feet, syllables, onsets, rhymes and segments. Because linguistic
code is hierarchically organised, linguistic theories assume that speakers’
minds must incorporate ‘rules’ not only for relating bits of code to bits
of meaning, but also for relating smaller bits to larger bits. Apart from
such ‘structure building’ rules, the need for assuming ‘structure changing’
rules has arisen as well, relating more superficial representations of code
to assumed ‘deeper’ ones, which may sometimes be different.
In short, linguistic theory provides us with a host of concepts for
analysing linguistic code and relating it to meaning. The above exam-
ples include merely a random selection, which does not even intend to be
representative of the vast descriptive repertoire that centuries of linguistic
practice have amassed. Its purpose is merely to indicate the vastness of
the task we are facing. For each linguistic category – be it simple, com-
plex, a rule, a bit of code or a bit of meaning – there exists the possibility
that it might represent a competence property that replicates.
Clearly, any attempt to consider even only a representative number of
them in detail and discuss how likely it is that they might actually be lin-
guistic replicators would merit a volume of its own. Therefore, we must
deal with the issue in an exemplary manner. That is to say, we have to
choose a small set of categories and see how far we get with them. Since
5 Importantly, one has distinguished between the material representation of code in writ-
ing, speech (and, recently, signing) and the representations of the code within minds.
Linguistics has typically focused on the latter.
Towards an evolutionary theory 127
which rules a competence model would need for combining them into
the larger, and ever new pieces of discourse that we observe. Therefore,
a typical strategy has always been to analyse discourse into smaller com-
ponents, and to devise empirically adequate and possibly powerful rules
for their (re-)combination. The approach will be familiar to everybody
from grammatical parsing and from spelling. As a result, the competence
models that linguistics has produced have thus come to contain primi-
tives which are on the small side: ‘phonemes’, ‘syllables’, ‘morphemes’,
‘words’, and so on. A competence which can deal with a sentence like
John is easy to please will normally not have to retrieve a copy of that
whole sentence from memory, but rather copies of the words John, easy,
to, please and be. It will have to know, additionally, which of them is a noun,
an adjective, a verb or a particle, and so on. Then it can use components
such as these to assemble different sentences such as John pleases Mary,
It pleases John that Mary is here, and infinitely many others.
The details differ from theory to theory, but what matters is that the
components of which linguistic competence models consist are typically
small. For our purposes this is good news, because the smaller a com-
ponent is, the more easily it will qualify as a replicator. This is because
the criterion of copying fidelity will disqualify as potential replicators
all configurations that do not seem to behave as indivisible units in the
assumed replication process. Therefore, competence properties which
have idiosyncratic distributions and which may vary independently of
others will not be parts of bigger replicators.
In order to see why this must be so, consider a few potential candi-
dates. Start with the biggest possible unit, that is, with the possibility that
whole competences might copy intact. If this were the case it would mean
that when a child acquires a language from her parents, she acquires the
complete copy of a competence ‘for’ English. Only then would ‘English’
deserve to be called a replicator. The criteria of stability and fecundity
would not rule this out: any ‘language’ is stably represented in the minds
of all speakers once they have acquired it. Also, there is sufficient evidence
that a sufficient number of speakers keep acquiring English to make up
for speakers who die. Again, this is trivially true of any language as long
as it ‘is spoken’. However, ‘English’ certainly does not pass the copying
fidelity test. As we have already mentioned no two speakers speak exactly
alike and not all inter-individual differences can reasonably be attributed
merely to performance factors.7
7 While it might be argued that some of these differences are irrelevant because they might
have causes that are not strictly speaking linguistic, this argument cannot be used within
the approach we have taken here. We have defined competence properties as properties
‘for’ particular types of behaviour, and when we observe differences in such behaviour
we shall have to accept them.
Towards an evolutionary theory 129
[. . .] I have talked about memes as though it was obvious what a single unit-meme
consisted of. But of course it is far from obvious. I have said that a tune is a meme,
but what about a symphony: how many memes is that? Is each movement one
meme, each recognisable phrase of melody, each bar, each chord, or what. (1989:
195)
He even suggests the very method we have employed here for identify-
ing actual memes, which is analogously applied in evolutionary genetics,
where a gene may be defined
8 In this respect, they might be beaten by lexical replicators, if there should in fact be any.
134 Selfish Sounds and Linguistic Evolution
point which disqualifies them as replicators. The fact that all speakers
seem to have them might suggest that their existence does not depend on
replication at all. Instead, the capacity of raising one’s tongue, the capac-
ity to create an egressive airstream, or the capacity to distinguish among
various types of acoustic signals are likely to be part of every human’s
genetic endowment. If these properties do indeed replicate, then it will
probably be genetically, not culturally or ‘memetically’. Thus, they are
unlikely to qualify as linguistic replicators proper. Instead, they seem
to constitute for language what nucleotide acids represent for biological
life, that is to say the low level building blocks from which larger units
emerge. It is only on such larger units, however, that natural selection
can come to act. Thus, phonemes are both small enough to copy intact
and complex enough to allow for variants to arise. Therefore, they qualify
better as replicators and possible units of selection than phonological fea-
tures do. We shall therefore include them in our tentative list of linguistic
memes.
haven’t? Does somebody ‘know’ the proper meaning of the word unless
she is aware that a bull belongs to the family bovis?
The last questions point to yet another issue that has been riddling sys-
tematic attempts to pin down and describe meaning. Meaning is by defini-
tion subjective. Words don’t just ‘mean something’, but they always mean
‘it’ to ‘somebody’. While the formal side of people’s linguistic behaviour
is comparably easy to describe in intersubjectively verifiable terms, the
‘meanings’ in their minds show up only indirectly in their behaviour. This
makes it extremely difficult to decide if a word really means the same to
any two speakers. This is particularly unfortunate since this is one of the
central questions that needs to be answered if one wants to know whether
meanings can be assumed to replicate faithfully and intact, and this would
clearly require a third-person perspective on what a morpheme means to
different individual speakers.
The issue is still further complicated by the complex interplay between
allegedly stored word meanings and contextual or pragmatic factors in
actual discourse. The meanings which words convey in the context of
use appear to be very diverse. How much of its alleged lexical meaning
‘to cause somebody to die’ does kill ‘carry’ in texts like the following, for
example?
(15) My sister would kill me (=be very angry with me) if she heard
me say that.
Lack of romance can kill a marriage.
The doctor gave her some tablets to kill the pain.
The two of them killed a bottle in one evening.
We were killing ourselves laughing.10
Does a word still carry all of its ‘proper’ meaning when it is used fig-
uratively or metaphorically? How often does a word have to be used
‘metaphorically’ for the metaphorical meaning to become its proper one?
Apart from well-known difficulties in the description of meaning, how-
ever, there is also a more fundamental problem, namely that semantic
descriptions like the ones just given, represent themselves only (meta-)
linguistic labels for assumed competence constituents. They suggest that
meanings are stored in minds in a way that is similar to language itself.
Some linguists, like Steven Pinker for instance, refer to that assumed
language as ‘mentalese’ and call it ‘the language of thought’. Clearly,
however, assuming that meaning exists in human minds in another form
of language is just a way of making it easier to label assumed concepts, and
conceals how unresolved the issue really is. Semantic labels like ‘[elk]’, or
relations between the constituents of ‘meaningful’ forms, while the semiotic or associa-
tive binding between the formal and the semantic sides of signs may be relatively loose. In
Deacon’s (1997) terminology, linguistic signs should be conceived of as ‘symbols’ rather
than ‘indexes’. Although his definition of the two terms is somewhat non-standard from
the linguistic point-of-view it makes good sense in the context of the present discussion.
While indexical relations involve a stable relation between a signans type and a type of
signatum (such as Baum ←→ ‘tree’), in symbols signantia are only loosely related to a
group of signata, while at the same time being (syntagmatically) related to a set of other
potential signantia (Baum might be related to a specific set of other words Ast – wachsen –
Rinde – Blätter – Zweige – Verästelung – Wurzeln – schneiden – Holz and so on. Furthermore,
as a ‘noun’ it might be associated to all ‘verbs’, to ‘determiners’, ‘modifying adjectives’
and so on). Which of the possible signata to which a symbolic signatum is associated gets
activated in individual semioses, among other things on the presence of other signata.
If this is true, it would make it easy to understand why structuralist semantics has had
immense difficulties trying to establish word meanings. At the same time, such a view
would account for the apparent context-independence and the ‘proactivity’ of human
language and, indeed, human thought, that is, the fact that we can talk and think about
things that are not present when we speak and think.
Towards an evolutionary theory 141
(16) Syllable
Onset
Rhyme
Nucleus
Coda
Phonemes /k/ / / /t/.
13 The literature on the syllable is vast and growing. For the present purposes being up-to-
date does fortunately not matter, but see for example Anderson (1986), Clements/Keyser
(1983), Dziubalska (1995, 1996), Giegerich (1985), Goldsmith (1976), Harris (1994),
Hooper (1972), Kahn (1976), Lass (1976), Murray/ Vennemann (1982), Hogg/McCully
(1987), Vennemann (1988), Vincent (1986).
142 Selfish Sounds and Linguistic Evolution
14 Like the syllable, the foot is a hotly debated subject in contemporary phonological theory,
so defining it in the rather straightforward manner of Abercrombie (1964) might seem
inappropriately simplistic to some. But I still find Abercrombie’s approach very much
viable. See also Dziubalska (1995), whose approach is compatible with the one taken
here, for further arguments.
Towards an evolutionary theory 143
might thus provide units for selection to choose among. Quite generally,
like all linguistic constituents, syntactic ones will also qualify as replica-
tors only if identical copies of them exist in a sufficiently large number of
competences while at the same time not being universal.
Since the empirical part in the later sections of this book will focus on
the evolution of phonological and morphological replicators, syntactic
questions will not be pursued much further here. Yet, one issue needs
to be addressed. As observed at the beginning of this section, mental
constituents for syntactic categories appear to differ from the morpho-
logical and phonological replicators which we identified earlier in that
their expressions represent a highly heterogeneous set of textual bits.
This difference, and in particular the fact that the textual expressions of
phonemes, morphemes and larger morphemic sequences tend to look
relatively similar to one another may cause a rather serious misunder-
standing.
Assume ‘N’ to be a mental syntactic category, a potential syntactic
replicator which gets expressed in discourse by elements such as man,
table, staircase and so on – that is, by any item that we would recognise
as a ‘noun’. All speakers competent in English must have a copy of that
particular constituent in their minds, even if they do not know that lin-
guists refer to it as ‘noun’, or ‘N’. ‘Having’ that constituent is what it
takes to use words that are nouns in the appropriate way, or to process
them accordingly when exposed to them. If a speaker does so, his/her
mind can be diagnosed to possess it. Assuming that such constituents
exist, what are they likely to look like? Being mental entities they must
clearly be realised, ontologically, in human brains. There, however, the
potentially replicating constituent that linguists refer to as ‘noun’ or ‘N’
will itself bear no such label of course. It will be the neural configuration
that is involved in the production and the cognitive processing of nouns,
and will ‘look like’ brain tissue. This means that the competence property
‘for’ nouns will bear no similarity whatsoever to either labels like ‘N’ or
‘noun’ or, indeed, to any of the particular nouns it ‘is for’. Of course,
this may strike you as self-evident. To assume that a neuronal constituent
involved in the processing of nouns should resemble stretches of text like
‘table’, ‘man’, or even ‘noun’ is as absurd as believing that genes for blue
eyes should be blue.
However, what strikes anybody as absurd in the field of syntax is not
so easy to recognise as wrong in phonology and even morphology. In
fact, even experts sometimes fail to do so. While individual instances of
nouns look very different from each other, and from category labels like
‘N’, or ‘noun’, this is not the case with speech sounds or morphs. For
example, the discourse realisations of a phoneme such as // are usually
148 Selfish Sounds and Linguistic Evolution
same way, or, as one might also put it, they will function in the same way
when interacting with minds. The phoneme itself will then be that config-
uration within a human mind which ensures that this indeed happens. On
the other hand, a ‘phoneme’ will be a mental configuration which under-
lies the production of sounds with the same effect or function. What we
mean by saying that a competence has a particular ‘phoneme’ is that it
will make its speaker perform a specifiable set of gestures under particular
conditions and react predictably and uniformly when exposed to one of
a specific class of acoustic impressions.
Conceiving of ‘phonemes’ in such a way is compatible with the notion
that they will – ultimately – ‘be’ neural configurations, a view to which
there really can be no serious alternative. Being neural configurations, of
course, phonemes can bear no similarity whatsoever to either the artic-
ulatory gestures performed when they are activated or to the particu-
lar acoustic patterns that result from this.15 Thus, the fact that we call
‘phonemes’ by similar names as sounds and articulatory gestures is a
matter of convenience with a very confusing side to it.
If one takes this into account, one will see that syntactic categories
are not different from categories such as phonemes or morphemes at
all. When we say that a competence includes the category ‘NP’, this
means that a speaker with that competence will react to a class of tex-
tual sequences (which may be as different as John, The house, Yesterday,
Whatever you say, The man I said had called earlier) in ways that have some-
thing in common (such as by knowing that they can either follow a verb
phrase or be followed by one). Similarly, the conditions under which a
speaker may produce any of such sequences of text will have certain fea-
tures in common as well. Thus, the label ‘NP’ refers to a particular way in
which speakers’ minds categorise their textual experience and modes of
behaviour, or to that configuration within a speaker’s mind by which such
categorisation is effected. In a speaker’s mind, in other words, the entity
referred to as ‘NP’ is that configuration which is responsible for the con-
sistency in his/her reactions and behaviour. Therefore, it is wrong to think
that the relation between a mental category ‘NP’ and the stretch of text
The man who lives next door should be different in principle from the rela-
tion between the mental category // and a bit of spoken text transcribed
as []. In both cases, specific textual patterns with many contingent quali-
ties trigger mental responses which could be equally triggered by different
textual patterns (such as [i] or [-i] in the case of //, or John, the house, my
15 The assumption that a phoneme might be a slightly abstracted or idealised ‘image’ of a
speech sound, passed on to some central processing unit in human brains is based on a
Cartesian view of consciousness as a kind of theatre in which pre-analysed perceptions
are displayed to a central observer, and is therefore fundamentally misconceived.
150 Selfish Sounds and Linguistic Evolution
cat etc. in the case of ‘NP’). If there is a difference then it will be that
The man who lives next door, when received, will trigger a complex set of
mental responses among which the identification of the stretch of text
as a ‘NP’ will be only one, while the mental response for which a sound
such as [] can be responsible may rarely go beyond the triggering of //.
Even more generally, we might say that there is no real difference
between any of the competence constituents that we have so far dis-
cussed. Basically, ‘phonemes’, ‘morphemes’, ‘syntactic categories’ as
well as semantic ‘concepts’ must all equally be understood as mental
constituents for recognising, representing and behaving appropriately
towards aspects of the environment in which humans live.16
Let us turn next to the final type of competence constituent whose
potential replicator status is to be discussed in this section, namely rules.
16 Now, one of the reasons why I have brought syntax up is that I shall not have to say much
more about it in the rest of this book. Mostly because of my own professional background
and expertise, the more detailed case studies I shall present will focus on phonological
and morphological phenomena. Since it is fairly obvious, however, that the perspective
on language and language change that is beginning to emerge must be relevant to all
recognised levels of linguistic description, the complete exclusion of some might be
experienced as a serious omission. The reason I included the above two paragraphs,
then, is that I have felt obliged to point out that there are syntactic aspects which require
further consideration, to suggest (though admittedly in a very sketchy manner) how
they might relate to the phonological and morphological cases which I shall discuss in
some detail, and to encourage further enquiries into the matter. Thus, although it might
disrupt the flow of the narrative a bit, I will continue to include digressions of a similar
kind in the next sections.
Towards an evolutionary theory 151
that speakers don’t memorise all the variant shapes that may express a
morpheme but acquire rules for deriving some from others. Ideally (in
terms of computational efficiency), each morpheme would have just a
single ‘underlying’ form, which is stored in (lexical) memory and from
which all variants are derived by rule. At the same time, of course, it is
desirable that the assumed rules should themselves have possibly wide
coverage, that is, explain as many morphophonemic alternations as pos-
sible. In the case of {was} one would therefore want to assume (among
others) a rule that relates [z ] and [s ]. In order to find the appropriate
version of such a rule, one will take into account that /s /-variants tend
to occur primarily when {was} is followed by a word that begins with a
voiceless consonant as in
(17) It was Tom. / twə[s]tɒm / vs.
It was Ivan. / twə[z]a vən /
One will also consider that the same type of [z ]-[s ] alternation occurs in
other morphemes as well, as in
(18) It is Tom. / t [s]tɒm / vs.
It is Ivan. / t [z]a vən /
It pleases Tom. / tpliz [s]tɒm / vs.
It pleases Ivan. / tpliz [z]a vən /
and so on.
All these things considered, one is likely to come up with some version
of an assimilation rule like
(19) C → [−voice] / [−voice]
Now, contrary to the constituents we have discussed so far, rules like
(19) seem to represent mental processes rather than static entities. This
may raise doubts about their stability, because while entities have spatio-
temporal integrity, processes normally don’t. So rules might not meet
the longevity criterion and fail to qualify, therefore, as linguistic repli-
cators. However, we are not asking whether mental processes are stable
but whether mental configurations are, which serve ‘for’ producing and
dealing with sounds whose occurrence in texts can be accounted for in
terms of processes. That is different. Such a mental configuration need
not itself be, and probably isn’t, dynamic at all, and that it should be long-
lived is, in principle, just as plausible as it is in the case of constituents
‘for’ other bits of text. Thinking of computer programmes is helpful here.
A programme may contain a rule in its code. As part of the code, the rule
is of course temporally stable, while only the computational events which
occur when it is actually invoked and changes the state of the central
152 Selfish Sounds and Linguistic Evolution
17 Take German final devoicing as a classical example which speakers normally find difficult
to suppress when acquiring languages such as English.
Towards an evolutionary theory 153
we have seen that some of them are indeed likely to qualify as linguistic
replicators. Among them are competence constituents for (a) phonemes,
(b) phoneme configurations (with characteristic suprasegmental structures)
that are associated with ‘meaning’, (c) foot types, as well as phonological
(d) rules. For all of them, we could assert that they are highly long-lived
in human minds which have acquired them and normally copy faithfully.
As far as syntactic categories and configurations are concerned, we
said that they are, in principle, also likely candidates, although we did
not commit ourselves on any particular cases. Similarly, with regard to
the semantic side of language, we argued that in some way conceptual
configurations are also likely to be stably represented in and faithfully
transmitted among human minds, but we did not commit ourselves on
specifics in this regard either.
As far as more complex associations of constituents from different
domains are concerned, we discussed two cases which we preferred
to regard as potential replicator alliances rather than as proper replica-
tors. The first case concerned morphemes in the sense of formal units
that carry meaning. There, we argued that conceptual replicators, what-
ever they may eventually turn out to be, are unlikely to be the same as
the constituents which are generally regarded as the ‘meanings’ which
morphemes can convey. We suggested that form–meaning pairings as
assumed in lexicography and traditional structuralist morphology may
have no status as units in a replicator based approach to language.
Instead, we suspected that the mental associations between morphotac-
tic forms and conceptual configurations may be too loose for qualifying
morphemes – viewed as units in which meanings are bound to forms – as
proper replicators, so that we preferred to regard them as looser replica-
tor alliances, or ‘memeplexes’. Something similar applies to the second
case we discussed, namely the relation between lexically stored phoneme
sequences, or morphs, and rhythmic units such as foot types. Like in the
case of morph-meaning pairings, we argued that the associations between
them might better be regarded as temporary alliances which do not seem
to be stable enough to qualify proper replicators, or ‘memes’.
6.1.6 Résumé II: mental replicators, how to keep them apart from their
extra-mental expressions, and why this is important
An issue which came up repeatedly and particularly with regard to syn-
tactic categories and rules, was the relationship between mental linguistic
replicators, or replicating competence constituents, on the one hand, and
their expressions in discourse and text on the other. For the approach
which is being developed here, it is important to keep the two neatly
apart. An evolutionary theory of language (like any theory that follows
154 Selfish Sounds and Linguistic Evolution
includes the phoneme /ɑ /, this ought to mean that this competence is
configured in a manner that will make its ‘owner’ or ‘host’ behave in a
specific way if exposed to one of a possibly rather mixed set of different
actual speech sounds (ranging, in variable contexts, possibly from [ə ] or
[ə ] over [ɒ ], [ɑ ], [a], [a
], to [a ], [æ
] [ε]), – a way which will be cate-
gorically different from the ways in which s/he behaves when exposed to
sounds outside the set. (Consider likely reactions to Take the car /kɑ /. vs.
Take the key /ki /.) In this sense, the notion of a ‘phoneme /ɑ /’ refers to
whatever in a speaker’s mind realises that configuration. It refers neither
to articulatory gestures, nor to sounds. Equally, it must not to be misread
as meaning that speakers’ minds actually host some idealised version of
[ɑ ], and it should be clear that while ‘having this phoneme’ may make
a speaker perform one of a set of articulatory gestures under specifiable
conditions, such as when the need to communicate about bras, cars, tar,
and so on, arises, it does not mean that an idealised version of this sound
is getting ‘realised’, or ‘expressed’, in the process. The competence prop-
erty /ɑ / is a mental constituent ‘for’ certain types of behaviour. It is in
no way ‘like’ that behaviour, nor ‘like’ its textual products.
The same is true of other categories of linguistic descriptions. If a lan-
guage is said to include a phonological rule like assimilatory consonant
de-voicing (C → [−voice] / [−voice]), this means that speakers will pro-
duce consonants without voicing them before unvoiced segments, even
though they will produce voiced ones in other contexts. Also, when they
hear unvoiced consonants in such environments they will distinguish, in
their behavioural reactions, between such consonants that they associate
with alternative voiced pronunciations and such that they don’t. To the
degree that such behavioural consistency is likely to have a mental basis,
it can be assumed that there will exist a common mental configuration
in speakers’ minds which underlies each behavioural event that looks
as if a ‘de-voicing process’ might be involved in it. Again, that mental
configuration is unlikely to look very much like the established linguistic
descriptions of such a process.
Or take another example. If a language is said to include, say, past
tense marking by means of a suffix {ed}, then this is meant to imply
that speakers will react in specifiable ways when they hear that suffix and
will produce it, likewise under specifiable conditions. It implies mental
constituents ‘for’ {past tense} and ‘for’ {ed}, as well as a mental config-
uration that relates the two. Neither of them will look, sound or feel like
‘the past’ or like the morpheme {ed}.
Confusing replicators and their expressions can have far reaching
effects. In genetics, it might tempt one to think that there could be a
one-to-one relationship between phenotype properties and genes. This
156 Selfish Sounds and Linguistic Evolution
(1866) experiments were digested, and long before genes themselves were
actually discovered. They had to assume units of inheritance, without
knowing what those units might be. All they saw were similarities between
phenotypes. Likewise, the linguistic assumption that competences which
seem to have similar effects will also have similar structures is hypothet-
ical. Since our approach depends on this hypothesis, we need to discuss
how plausible it really is.
one person’s brain hosts a copy of a linguistic constituent that also exists
in other brains would be utterly meaningless.
Clearly, this situation is somewhat uncomfortable, and it would be nice
if there were some easy way out of it. It would be great, for instance, if
there were evidence that linguistic replicators existed on levels that are
more easily accessible to empirical observation. Although all we have
observed so far speaks against this possibility, we should perhaps give it
some further thought. One linguist who has done so is William Croft
(2000). Similarly inspired as we are by the explanatory strategies that a
generalised theory of evolution seems to offer, he has proposed an evolu-
tionary model of language change which is based on the idea that ‘utter-
ances’ should be linguistic replicators. Clearly, Croft’s proposal appears
attractive at first, because it appears fairly easy to determine, by simply
comparing the physical properties of utterance stretches, whether two of
them are similar enough to count as copies of one another. However, it is
flawed in a fundamental way. This becomes clear as soon as one considers
how Croft defines an utterance. According to him, it represents
an actually occurring piece of language, completely specified at all levels of struc-
ture, including its full contextual meaning on the particular occasion of use (i.e.
speaker’s meaning). (2000: 244)
things. It certainly does not mean that the latter have no neuronal basis. It
is much more plausible to assume, instead, that higher-level mental con-
stituents will turn out to correspond to larger and internally structured
assemblies of – possibly locally distributed – neurones. That neurones
tend to form assemblies seems to be well established. This happens when
a set of them happen to fire simultaneously, or in close temporal suc-
cession. Then, they will form higher-level constituents which always fire
in unison, almost as if they were one big cell rather than many indepen-
dent ones.22 It was first suggested by Donald Hebb that such assemblies
might represent more probable candidates than individual neurones for
the ‘much-needed bridge between the structures found in high-level cog-
nition and the nervous system’ (Anderson 1995: 285). This idea is still
taken seriously among cognitive scientists, although it has so far with-
stood experimental testing.
Whatever the exact format in which brains store and represent infor-
mation may be, however, the key to this capacity will probably be that
they form dynamic networks, and that these networks can adjust their
own structures in response to environmental feedback. Two pieces of
evidence speak for this. First, the networks represented by neurones are
indeed capable of adjusting the qualities of inter-neuronal connections,
thus channelling the internal flow of electro-chemical energy into specific
pathways. Second, attempts to simulate learning behaviour in comput-
erised models of networks with adaptable connection strengths among
individual nodes have been highly successful. All this clearly supports
the hypothesis that cognition, learning and memory are a matter of asso-
ciation.23 The particular ways in which the internal organisation of a
brain comes to be adjusted are likely to reflect environmental feedback
on the behavioural effects of variant states. It is possible that such feedback
is transmitted to the mental network in the form of neuro-transmitters
under the control of the limbic system, a part of the brain which is likely
to play a role in emotions (Plotkin 1994). Simplifying a lot, a mind
that assumes functional states seems to be ‘rewarded’ by ‘feeling good’,
and one that assumes a dysfunctional organisation punished by ‘feeling
bad’. Configurations which incur positive feedback are thereby stabilised,
22 In fact, the study of complex systems suggests that the emergence of higher-level struc-
tures through auto-catalytic self-organisation is almost to be expected if such systems
contain a sufficiently large number of richly interconnected constituents (see, for exam-
ple, Kaufmann 1995: 54–69).
23 The notion that memory should be associative has a very long tradition. One of the first
modern psychologists to suggest that the relevant associations might actually involve
excitatory connections among ‘point[s] in the brain-cortex’ was William James (1892:
226), and the most influential advocate of this idea in the twentieth century was Donald
O. Hebb (1949), whose views underlie most contemporary connectionist approaches to
modelling learning and cognition.
162 Selfish Sounds and Linguistic Evolution
B
C
Figure 6.1 How to identify constituents in structured networks.
B
Figure 6.2 How to identify copies of network constituents.
A C
A
B
D
C
constituents, and to determine how similar they are to one another. Sec-
ondly, these arguments will necessarily hold regardless of the specific
material level on which network structures are implemented in human
brains, as long as brains represent, store and process information in terms
of network structures at all. It will make no difference if, in practice, that
should be on the neuronal level, a level of neuronal assemblies, or a level
of other and internally more complex constituents.
Now, if the way in which brains store information can be described in
terms of relational networks, then this it implies a level on which ‘function’
(that is, the representation of knowledge, or behavioural options) meets
‘form’ (that is, the internal and external structures of node clusters). On
this level, constituents ‘for’ similar types of knowledge and/or constituents
of behaviour will be formally similar, that is, occupy similar relative posi-
tions and have similar structural characteristics. Network constituents
‘for’ similar knowledge and similar behavioural possibilities will thus also
be structured similarly. If they are sufficiently long-lived, and transmit-
ted at a sufficiently high rate and a sufficiently high degree of fecun-
dity, they would be perfect as candidates for the role of mental/cultural
replicators or ‘memes’ and might represent the material basis which
an evolutionary theory of culture in general and language in particular
requires.
At the same time, if memes are understood as internally structured
node assemblies embedded within larger networks, it becomes possible
to see in what sense two different instances of them could count as com-
peting ‘variants’ of one another. If the identity of memes is established
both through their internal make-up and through their position within
larger network structures, it is obvious that a single ‘position’ within a
network may be ‘occupied’ by more than one type of assembly configu-
ration, or meme. Thus, the B-constituent in (figure 6.3) is structurally
different from the B-constituents in figure 6.2) and (figure 6.1), and the
way in which both B-‘types’ relate to A and C makes them ‘variants’ of
each other. Since the frequency of one B-type in a population of net-
works will correlate negatively with the frequency of the other, they can
be conceived of as ‘rivals’ or as competitors for a specific slot within the
network. Thus, they would qualify as memetic counterparts to biological
‘alleles’, that is, different genes in competition for a single position on a
genome.
6.2.3 Summary
We have now sketched an assumption of what the material basis of
‘memes’, that is, mental replicators, linguistic or otherwise, might ‘look
Towards an evolutionary theory 169
like’. While its point has not been to develop a definite position on the
question how brains encode and process information,27 it has shown that
there can, in principle, be a level of brain organisation on which units of
information are implemented in ways that give them formal structures –
even though it might be unclear at present what level that will turn out
to be. The existence of such a level then suggests the following defini-
tion of a ‘meme’, which we shall tentatively adopt during the rest of this
book.
(20) A ‘meme’ represents an assembly of nodes in a network of neu-
rally implemented constituents, which has (a) a definite internal
structure, (b) a definable position within a larger network con-
figuration, (c) qualifies as a replicator in Dawkins’ sense.
On this definition it is furthermore possible to regard meme variants
which can occupy the same position within types of network configura-
tions as being in competition for that position. This provides a suitable
basis for studying memetic evolution in terms of variation being brought
about by imperfect meme replication and the subsequent selection of
competing variants, as the Darwinian paradigm requires.
27 It does not ‘advocate’ any of the specific connectionist models of (aspects of) linguistic
competence that have been proposed in the wake of Rumelhart/ McCleland (1986), but
is definitely highly sympathetic to the research programme. For a good overview over
connectionist achievements see Spitzer (1996), and for a defence of the approach see
the debate in McDonald/McDonald (eds. 1995). But see also Pinker (1999) or Dressler
(1999) for justified criticism of ‘greedy’ connectionism.
170 Selfish Sounds and Linguistic Evolution
6.3.1 Phone-memes
As ‘memes’ for categorising, representing, producing and interpreting
speech sounds, phone-memes will have to incorporate constituents that
(a) respond to and generalise over auditory impressions and that (b)
direct the performance of articulatory gestures. They must be linked,
externally, to constituents for recognising, representing and processing
phonemic sequences that represent the Gestalts of morphemes. Thus, the
‘phone-meme’ for English /z /, as in was, zoo, busy, goes, etc. might be
conceived of in a way similar to figure 6.5: 28
Thus, the phone-meme /z / amounts to an association between a com-
plex of specific articulatory gestures, a specific auditory impression as
well as with a set of morph-memes in whose identification /z / makes a
difference. Of course, the representation above is fragmentary and sim-
plified in many respects – not only with regard to neuronal structures,
but also with regard to the specific articulatory and auditory features it
should contain. About the neuronal aspect I know far too little to com-
mit myself, and as far as particular phonetic features are concerned, they
could clearly be discussed at great length, but their exact nature would
not contribute a great deal to the present argument. What matters here
is that a phone-meme can plausibly be conceived of as a configuration in
which mental constituents ‘for’ auditory impressions are linked to mental
constituents ‘for’ performing ‘articulatory gestures.
A few things deserve to be clarified. First, the representation must not
be conceived of as a diagrammatic blueprint of actual neural constel-
lations. As already said, it is merely supposed to demonstrate that the
functions which phonemes play can in principle be materially imple-
mented in terms of network-like structures. The particular structure
below, however, is unlikely to bear any resemblance to actual structures
as may eventually be identified in actual human brains. For example, it
28 Note that the labels in the diagram are there only for the sake of convenience. They
indicate the functions of constituents, but have themselves no neuronal status in speakers’
minds.
Towards an evolutionary theory 171
... ...
... ...
/z/
Phone-Meme
{was}
{zoo} {-es}
{busy}
{...} {...}
Morphs
29 In terms of the theory of neural networks, it may correspond to a node on one of the so-
called intermediate layers, whose constituents are involved in forming abstractions of the
very type that ‘phonemes’ clearly represent (cf. Spitzer 1996: 129–32). It is thus merely
a part, albeit a crucial one, of a constellation that produces cognitive and other behaviour
‘for’ /z /. Thinking of it in terms of a ‘mental /z /’-node, however, would nevertheless be
misleading, because mental /z /-ness is not locally encoded in the node but distributed
over a larger section of the network.
172 Selfish Sounds and Linguistic Evolution
and that we have labelled it /z /, is meant to make things easier for read-
ers. Actual ‘/z /-ness’, or ‘knowledge of /z /’ is a function that emerges
from overall link patterns and from the ways in which constituents
interact.
Next, it is important to distinguish between the structure of a phone-
meme, which can be thought to implement ‘knowledge of a phoneme’
and its behaviour in ‘expression’ or ‘activation’, that is, when a phoneme
is either perceived or produced. For instance, not all of the structural con-
stituents of a phone-meme will be ‘active’ in every instance of language
use involving it. If that were the case, one would automatically produce
a phoneme whenever one hears or thinks of it, and nobody does that, of
course. Therefore, selected subsets of either articulatory or auditory con-
stituents will remain inactive even though the phone-meme is involved
in activity in one way or the other. When one whispers, for example, the
activation of articulatory voicing is systematically suppressed, and there
are many other conceivable cases by which the principle could be illus-
trated. In short, ‘knowledge’ of a phoneme can certainly be thought of as
the configuration of nodes and links which constitute a ‘phone-meme’,
but such a unit of knowledge is not identical with a unit of expression or
activity. Rather, it defines a web of pathways which make some patterns
of activation more probable than others. Thus, phone-memes represent
(comparably) stable competence constituents, that is, elements of ‘knowl-
edge’, but they will certainly not always operate as coherent units in their
expressions in actual discourse.30 And, in principle, this must be true of
all other memes as well, of course.
Before turning to morph-memes, consider the following interesting
aspect of the way in which we have represented phone-memes. Struc-
turally, they are characterised through the ways in which their constituents
are linked as well as through their ‘external’ links to ‘morph-memes’.
Now, if ‘phone-memes’ are indeed selected for – ‘rewarded’ and ‘sta-
bilised’ – during language acquisition, then there are two ways of looking
at this process. On the material, or structural level, one might think of
such ‘selection’ as a process in which a particular configuration of neu-
ral assemblies gets stabilised through being rewarded for firing together,
because this strengthens internal links. That they come to fire together at
all, however, reflects their common connection to neural assemblies ‘for’
30 This implies, quite generally, that the elements and processes involved in language use
do not necessarily have be identical with the elements and relations that define linguis-
tic competence and makes it improbable that views which model discourse as a pro-
cess in which competence constituents are put together to yield utterances will ever be
successful.
Towards an evolutionary theory 173
6.3.2 Morph-memes
Next, turn to memetic configurations for recognising, storing and pro-
cessing the formal Gestalts of morphemes. As we observed above, such
‘morph-memes’ will have to incorporate, first, stable and good links to
phone-memes as well as to memes for the suprasegmental roles played by
the individual phone-memes. Additionally, there must be morph-meme
specific links that handle the sequential, or temporal, organisation of the
involved memes for sounds and suprasegmental roles. Finally, a morph-
meme will be linked, ‘externally’ and by lines of diverse qualities to a
variety of different ‘syntactic’ and ‘conceptual’ constituents. A mental
constituent for a simple lexical morpheme like bull might thus be drawn
as shown in figure 6.6.
Reflecting what was argued above, the diagram in figure 6.6 shows
morph-memes as configurations in which phone-memes and constituents
‘for’ supra-segmental patterns and roles are stably connected. It is hoped
that the graph speaks well enough for itself, so few words will be spent
on its interpretation. What might require an explanation, however, are
the arrow symbols, because they did not figure in the representation of
‘phone-memes’. The connections they indicate are supposed to imply
sequential rather than concomitant firings. Nodes for syllabic constituents
are thus assumed to be linked in such a way that the activation of a
node ‘for’ onsets will cause the subsequent activation of a node ‘for’
nuclei, or at least make it more likely. More will be said about such
constellations below. In most other respects the diagram can be read just
like the representation of /z / in figure 6.5.
174 Selfish Sounds and Linguistic Evolution
Phone-memes
/υ/ /l/
/b/
‘Onset’ ‘monosyllable’
NOUN Morph-meme
[STRONG]
[ANIMAL] [BIG]
[DANGEROUS]
[HORNS]
Concepts [EDIBLE]
Legend: stands for memes which are themselves complex but whose internal structure
is not represented in the graph;
stands for functionally simple nodes, or nodes whose internal structure is of
no concern to the present discussion.
Figure 6.6 The morph-meme {bυl}.
/k/
/t/ /a/
/p/ /υ/
/s/
/ə/
/l/
/n/ /e/
Figure 6.7 A meme for distinguishing between onsets and nuclei.
/s/ O
/t/ O1b /a/
O1a
/b/ //
O1c O2a
O2b
/p/ /υ/
O3b,2c
/ə/
/l/
/e/
/k/ /r/
Figure 6.8 A meme ‘for’ the phonotactics of onset clusters.
(that is, those that can ‘play the role’), and because it feeds energy
(again through a special type of link) to a node which connects to a
different set of sound-nodes (namely those that can play nucleus roles).
The same kind of relational definition is possible, by analogy, for nuclei
and codas, of course, and by a similar rationale, to other constituents,
such as configurations ‘for’ consonant or vowel clusters, rhymes or
syllables.
Thinking of supra-segmental constituents in terms of specific link pat-
terns by which nodes are connected to one another and influence the
probability of one another’s activation, makes it possible to capture com-
plex phonotactic relations. Although it would clearly go beyond the scope
of this little exposition to design a network for the phonotactics of any
particular language, an exemplary demonstration of the principle seems
nevertheless to be in order. Imagine, then, a simple hypothetical language
that admits only the following onset types: (a) any single consonant, (b)
/s / followed by any voiceless stop, or liquid (for example, sk, st, sp, sl, sr),
(c) stops followed by liquids (tr, kr, pr, br and tl, kl, pl, bl), and (d) clusters
of /s / and voiceless stops followed by liquids (str, skl, spl, spr, and so on).
A network that would define the three onset positions and account for
possible relations might then look similar to figure 6.8.31
This network defines possible consonantal roles within complex onsets
in a distributed and relational manner. For instance, the fact that all
31 Which – for the sake of surveyability – does not include all the nodes and links necessary
even only for this simplified language. But the principles ought to become obvious.
Towards an evolutionary theory 177
32 Note how the network also seems to define natural classes of consonants, which shows
how membership in a class such as ‘liquids’ or ‘stops’ correlates with the suprasegmental
roles that individual phonemes can play.
Incidentally, this way of thinking about mental configurations ‘for’ syllable types is
highly compatible with linguistic approaches that do not regard ‘syllables’ as phonological
primitives, most notably with Dziubalska’s (1995 and 2002) beat-and-binding model
of phonotactic organisation. She argues that the structures which we are used to call
syllables emerge from mutual attractions (bindings) among vocalic phonemes (‘beats’,
in Dziubalska’s terms) and consonantal phonemes (non-beats).
178 Selfish Sounds and Linguistic Evolution
Nuclei
Onsets
Codas
‘monosyllabic foot’
Figure 6.9 A meme cluster for syllable structure.
/s/ N1
O1
C1
//
O2 /t/ /p/
/r/
O3 {strp}
/b/ N1
O1
C1
// /ə/
/t/
/v/
{bt} {of}
Figure 6.11 How {bit} fails to trigger C1 when occurring before {of}.
This indicates how syllabic structure can be more than just an epiphe-
nomenon of sequential arrangements of morphs or phonemes. The links
that define syllabic roles form structures with emergent properties of
their own. This possibility has many theoretical implications, but what
matters most for the present discussion is that configurations which
embody knowledge of prosodic relationships such as the ones expressed
in syllable structures ought, in principle, to be capable of replication in
their own right, and therefore represent integral ‘memes’ by themselves.
This is an important point because one usually tends to think of syllabic,
or rhythmic structures as properties which morph sequences have when
they are expressed in utterances, and it is therefore difficult to imagine
that knowledge of syllable or foot types should involve independent pieces
of information at all. (Can you think of a particular syllable without think-
ing of particular sound sequences at the same time?) One might therefore
find it difficult to imagine how ‘memes’ for syllabic or rhythmic structure
could represent independent units of replication. The diagrams in this
section help to make this more easily conceivable, it seems to me.33
6.3.3.2 Feet
Let us turn to feet next, English feet that is to say. In order to deal with
them successfully, a competence must be able, it would seem, to (a) distin-
guish between prominent syllables and weak syllables, (b) expect them to
alternate, (c) recognise sequences of one strong plus a variable number of
weak syllables as units of timing, and (d) calculate the probable duration
of feet on the assumption that the time span between prominence peaks
tends to be constant. If we think of competence being implemented as a
network, these tasks might call for structures of the following types: (a)
requires nodes ‘for’ recognising and producing different degrees of promi-
nence, each of them linked first to gestures for increasing and decreas-
ing articulatory effort respectively, secondly to perceived decreases and
increases of auditory intensity, and thirdly to nodes or assemblies for
those phonemic configurations (‘syllables’) with which the two degrees
of relative prominence get to be co-expressed; (b) requires the two nodes
to be linked in such a way that activation of one makes activation of the
other more likely; (c) and (d) require links between the nodes for strength
33 Of course, the model we have sketched is beautifully compatible with most contem-
porary theories of phonology, which treat syllabic structure as something that morph-
sequences ‘get assigned’ by syllabication rules rather than as something they ‘have’. At the
same time, it helps one to understand the apparent paradox that un-syllabified morph-
structures are impossible to even introspect. Since our model sees morph-memes as
being necessarily associated to memes for syllabic constituency, it is impossible to acti-
vate morph-memes without activating some syllabic structure at the same time.
Towards an evolutionary theory 181
S w
S w
σ2 σ3
σ1
/m/ /e/ /ə/ /r/ //
{memər}
O N C
Legend: 1 , 2 and 3 represent subsequent firings of what way be one
and the same neuronal mode. The three events have been graphically
separated in the chapter for purposes of perceptibility.
Figure 6.13 The rhythm of memory.
feet. What this suggests is that memes for foot types may be more inde-
pendent of memes for morphs than memes for syllabic roles appear
to be.
6.3.4 Rule-memes
The discussion of possible ‘resyllabifications’ and ‘rhythmic restructur-
ings’ in the last two subsections has highlighted an issue which we have
already touched upon earlier. It is so generally important that it calls
for a principled discussion. It involves the question of how a network
state might be able to handle the dynamic processes of speech produc-
tion and reception, and the structural transformations that they seem to
bring with them. As observed in section 6.1.4.5.2 above, many theories
of linguistic competence represent constituents for processes in terms of
‘rules’. The case we mentioned as a typical example was the phonologi-
cal assimilation of /z / to /s / before voiceless consonants as in It was [s t]
Tom. This process is often handled in terms of transformation rules like
C → [−voice]/ [−voice], which are believed to be activated in produc-
tion. Interestingly, the way the results of such processes are ‘handled’ in
perception is rarely formalised. Often it is implied that speakers will be
able to handle them in some way, simply by knowing the production rule.
Towards an evolutionary theory 183
The details are left vague however.34 While rules of the established for-
mat may work well in serially organised computer programs, however, it
is highly unlikely that actual minds/brains should implement information
‘for’ handling apparently ‘rule-governed’ relations such as those between
[s] and [z] in the above example in this manner – at least if brains are
really organised in terms of a network structure, as the most plausible
guess seems to be at the present time.
How can knowledge of a dynamic process be implemented in a static
network then? The problem is not as difficult as one might first imagine.
Take the example of /z /-devoicing in It was Tom again, and recall, first,
what kind of behaviour the network is supposed to generate. In produc-
tion, we expect it to put out [wə s] as the expression of the morph-meme
{was}, and in reception we expect it to activate concepts normally associ-
ated to the morph-meme {was} when it gets the phonetic input [wə s].35
In accordance with the way in which we represented the morph-meme
{bull}, we must assume that the morph-meme {was} incorporates a link
to the phone-meme /z /, which is supposed to activate [voicing] in pro-
duction. Activating [+voicing] automatically rules out the activation of
[−voicing] and vice versa, of course.36 How, then can [−voicing] be acti-
vated? Clearly, it must receive its input from some other source, and the
likely candidate is the /t / of Tom. In order for the network structure to
inhibit the prior activation of [+voicing] in It was Tom, little more seems
to be required than that the link between /t / and [−voicing] be so much
better than the link between /z / and [+voicing] that the energy emitted
by /t / will reach [−voicing] as soon as, or possibly sooner than, the energy
emitted by /z / reaches [+voicing], even though /t / itself might have been
activated after /z /. A graphic representation of such a network is given in
figure 6.14.
34 This may be because the problem is trickier than one might think at first sight. In our
specific case, for example, a receptive rule that voiced voiceless consonants before voice-
less ones will not work, because it would falsely interpret Jane wants to kiss Tom as /d
e n
wɑnts tə k z tɒm /.
As we have said, this is probably because, in the wake of generativism, linguistic com-
petence has typically come to be thought of as a production system, a piece of mental
software for translating ideas, concepts, propositions or whatever into speech. Inspired
by the format of classical computer programs, the competence models constructed on
this conceptual basis naturally came to contain constituents such as variables and bat-
teries of ordered transformational rules, of which C → [−voice]/ [−voice] is of course
a prototypical example.
35 For this we have, albeit partly indirect, behavioural evidence. There is no substantial
evidence whatsoever, on the other hand, for the assumption that anywhere in a human
mind a mental version of /z / gets replaced by a mental version of /s / (and vice versa).
Instead, this assumption rests entirely on an analogical transfer of rule-formats from
logical modelling, or computer programming.
36 This implies a mutually inhibitory connection between the two constituents.
Vocal folds: [-voicing] [-voiced]
Tongue: [corona towards alveoles] [+strident]
{wɒz}
{tɒ m}
sketched, it becomes obvious that there need not be a distinct ‘rule’ unit
on the level in which ‘knowledge’ of such a process ‘sits’, or ‘is repre-
sented’. There is no mental version of the rule, as in computer code. No
constituent of the configuration, which is ‘for’ the process, belongs to it
exclusively. All of them figure in other constituents and processes as well.
Yet, the ‘rule-meme’ does have an identity of its own, which is constituted
by the ways in which the various nodes that make it up are connected to
each other.
6.3.5 Summary
Since there is still much to be learnt about the processes by which brains
generate cognitive content, it is more than likely that the meme graphs
which I have drafted will turn out to be utterly unrealistic. Yet, they have
served to substantiate, I hope, the following points. First, it is not incon-
ceivable that the cognitive, or mental functions which brains serve might
have a material basis which can be described in structural terms. Second,
a systematic correlation or isomorphic relationship between patterns of
mental functions and patterns of neuronal structures ‘for’ functions can
plausibly be assumed. It would seem to follow from this that memes as we
have defined them, that is, as replicating neuronal structures with spe-
cific functions, expressions or effects, may indeed be as materially real
as the patterns of nucleotide acids that constitute genetic replicators are,
albeit possibly more difficult to observe and describe. This would then
corroborate the assumption that languages acquire, keep and therefore
have their properties because they evolve in a Darwinian manner. These
implications are exciting enough to justify, at least to a degree, the spec-
ulative way in which I have tried to give graphic shape to entities whose
exact material status is, let me stress that again, beyond our knowledge.
A A B B B
B B
t1 t2 t3 t4 t5 t6 t7
Nor do the details of selection matter when it comes to reconstructing
genealogies, or to charting family relationships that may come about when
different subsets of replicator populations get exposed to different selec-
tion pressures, and lineages split in the process. The tree in (22), for exam-
ple, can be established without knowing the reasons why or the means
by which competing variants came to be selected for and/or against. It
is enough to acknowledge the plain fact that selection seems to have
occurred.
188 Selfish Sounds and Linguistic Evolution
(22) A
Selection
A B for ‘B’
Selection
for ‘A’
B C
Selection Selection
for ‘B’ for ‘C’
happens among replicators will not make for an interesting story. In order
to explain the history of replicator populations it will be sufficient to state,
in a lapidary manner, that they respond to environmental changes. Then
one will have to focus on the latter.
Of course, the two alternatives just described are not mutually exclu-
sive. Obviously, situations are conceivable in which evolutionary changes
in a replicator population are partly driven internally, that is, through
the cumulative selection of replicators that provide an ever-increasing
fit between the pool and its environment, and partly externally, that is,
through environmental changes that redefine what constitutes such a bet-
ter fit. Furthermore, even the distinction between internal and external
factors will not always be easy to draw. After all, the different replicators
within a pool represent environments for each other. Even though it may
not be a matter of black-and-white but of darker and lighter shades of
grey, however, the distinction between externally and internally driven
evolution is important. The gist is that when one wants to understand,
rather than just chart or describe, the history, or evolution, of a popu-
lation of replicators, it does matter how the selection processes they are
subjected to come about and particularly, how stable they are. There-
fore, the mere assertion that constituents of linguistic competence seem
to qualify as replicators which copy imperfectly and are subject to selec-
tion, will not get us very far. Instead, we need to understand how their
replication is brought about and what factors may influence it.
Recall the situation that obtains in biological evolution as Neo-
Darwinian theory sees it. Actual evolutionary changes have clearly been
driven both externally and internally. What Darwinian theory focuses
on, however, is the logic of those processes in which replicators them-
selves can be considered active. Naturally, the course of biological evo-
lution on our planet has repeatedly been redirected rather radically by
catastrophic external events (such as the well publicised meteor impact
which assumedly extinguished all dinosaurs except birds and paved the
way for mammals). Yet, for much of the time selection pressures and the
resulting directions of evolution seem to have reflected the following state
of affairs: new replicator variants emerged and turned out to be better
than others under environmental conditions which, from the point of
view of the evolving populations were more or less constant. Under such
conditions, evolutionary change has tended to be cumulative and adap-
tive, and produced more and more sophisticated and satisfactory matches
between replicator populations and their environments. Therefore, in the
words of Richard Dawkins, Darwinian biological evolution can be com-
pared to a slow climb up the peaks of a (metaphorical) ‘fitness landscape’
(one of Dawkins’ books is called Climbing Mount Improbable). Normally,
190 Selfish Sounds and Linguistic Evolution
First and obviously, it raises the question in what medium, if not bodies
and brains, selves or souls are supposed to exist. This question cannot be
answered in any scientifically satisfactory way. It implies that there exists
a yet unknown ontological domain, for which there is no other motivation
than the very assumption that selves or souls are separate from brains and
bodies. It therefore epitomises an ad hoc notion which is grounded in the
very assertion that it is supposed to support.
If, on the other hand, the entities which we think of as human selves
or souls, are realised in brains and bodies, then they are on a par with
linguistic competence constituents, or language memes. Like the latter,
they must be neuronal configurations themselves. Being neuronal con-
figurations, of course, their internal structures and their places within
neuronal networks can in principle be determined in two ways. They can
either be genetically determined (programmed to become) hard-wired,
or they can be environmentally conditioned, that is, result from pro-
cesses of neuronal self-organisation directed by environmental feedback
on their effects. If such environmental feedback comes from neuronal
configurations in other brains and causes neuronal self-organisation to
produce copies of them, then the configurations which are thereby created
can count themselves as replicators. They also represent culturally trans-
mittable units of information, or memes, albeit of course not linguistic
ones.
What does this imply, then, for the role which ‘selves’ can play as envi-
ronments which select linguistic competence constituents? To the extent
that mental configurations ‘for’ human selves are genetically determined,
or hard-wired, the constraints on memetic transmission and evolution
that they represent will clearly be constant enough to allow the cumulative
selection of better adapted variants. The ‘fitness landscape’ they define
will clearly make it possible for memetic evolution to find directions.
Under such conditions, of course, the Saussurean notion that linguistic
change must be chaotic and unexplainable is not tenable. To the extent
that constituents of selves are transmittable, on the other hand, they may
themselves qualify as Dawkinsian memes – in very much the same way as
constituents of linguistic competence do. They will themselves have to be
approached accordingly. That is to say, changes in populations of memes
for ‘selves’, their ‘identities’, ‘tastes’ and other properties will have to be
explained in terms of selectional pressures against which fitter variants
will replicate better than less well adapted ones. Whatever those pres-
sures may be, whimsical human selves with randomly altering tastes and
preferences can no longer be among them.
This means that the Saussurean notion can be rejected on all accounts.
The view that human ‘selves’ will exert unexplainable and rapidly
194 Selfish Sounds and Linguistic Evolution
6.5.1 Introduction
Contrary to genes, memes – linguistic or otherwise – are not replicated by
a straightforward template copying process. Brains and their constituents
do not interact directly. Instead, memes are copied through a process
that is often referred to as imitation (see, for example, Blackmore 2000).
Although the agents which ‘do’ the imitating are normally supposed to
be ‘humans’ we have seen that this way of talking raises more questions
than it answers. The idea of human selves as autonomous overseers of
memetic replication is unfounded and likely to obscure the issue hope-
lessly. Instead, one needs to think of ‘humans’ holistically (and materi-
alistically) as the totalities of their minds and bodies. How then do the
brainy organisms that are humans manage to transmit acquired neuronal
structures among individuals? How can they ‘imitate’ what they do not
‘see’? And how, in particular, should they manage to produce sufficiently
faithful copies?
6.5.2 How can one copy what one cannot see? Revisiting
the ontological problem
It might be best to start by asking what humans (in the sense of organ-
isms with brains) do have immediate access to. To this question there are
reasonable answers. What humans perceive are contextualised instances
of human behaviour and its external consequences, including the imme-
diate results or products of that behaviour, as well as some of its further
consequences. In the case of language, they perceive how other people
speak, the textual products of such speech acts, as well as what is achieved
Towards an evolutionary theory 197
keep things as simple as possible [. . . and to] use the term ‘meme’ indiscriminately
to refer to memetic information in any of its many forms; including ideas, the
behaviours these brains structures produce, and their versions in books, recipes,
maps and written music. As long as the information can be copied by a process
we may broadly call ‘imitation’, then it counts as a meme. (66)
While this may be an option when one intends to introduce and pop-
ularise the idea of replicator based cultural evolution as well as its philo-
sophical implications, it is of course out of the question for our specific
purposes to keep things as informal as this. After all, there are a number
of theories about linguistic change which are technically highly sophisti-
cated and with which the present model needs to compete. Also, all that
is known about language shows that the cognitive level is clearly more
fundamental than the others, so that the level problem as such does not
really pose itself anymore. And this is exactly what makes the question of
how cognitive units can be replicated so urgent, of course.
Suppose we assemble a line of children. A picture of, say, a Chinese junk is shown
to the first child, who is asked to draw it. The drawing, but not the original picture,
is then shown to the second child, who is asked to make her own drawing of it.
The second child’s drawing is shown to the third child, who draws it again, and
so the series proceeds until the twentieth child, whose drawing is revealed to
everyone and compared with the first. Without even doing the experiment, we
know what the result will be. The twentieth drawing will be so unlike the first, as
to be unrecognisable. Presumably, if we lay the drawings out in order, we shall
note some resemblance between each one and its immediate predecessor and
successor, but the mutation rate will be so high as to destroy all semblance after
a few generations. A trend will be visible as we walk from one end of the series
of drawings to the other, and the direction of the trend will be degeneration.
Evolutionary geneticists have long understood that natural selection cannot work
unless the mutation rate is low. [. . .] How then can the meme, with its apparently
dismal lack of fidelity, serve as a quasi-gene in any quasi-Darwinian process?
[. . .] Suppose we set up our Chinese Whispers Chinese Junk game again,
but this time with a crucial difference. Instead of asking the first child to copy a
drawing of a junk, we teach her, by demonstration, to make an origami model of
a junk. When she has mastered the skill and made her own junk, the first child
is asked to turn round to the second child and teach him how to make one. So
the skill passes down the line to the twentieth child. What will be the result of
this experiment? [. . .] I have not done it, but I will make the following confident
prediction, assuming that we run the experiment many times on different groups
of twenty children. In several of the experiments, a child somewhere in the line
will forget a crucial step in the skill taught him by the previous child, and the line
of phenotypes will suffer an abrupt micromutation which will presumably then
be copied to the end of the line, or until another discrete mistake is made. The
end result of such mutated lines will not bear any resemblance to a Chinese junk
at all. But in a good number of experiments the skill will correctly pass all along
the line, and the twentieth junk will be no worse and no better, on average, than
the first junk. If we then lay the twenty junks out in order, some will be more
200 Selfish Sounds and Linguistic Evolution
perfect than others, but imperfections will not be copied down the line. [. . .]
The twenty junks will not exhibit a progressive deterioration in the way that the
twenty drawings of our first experiment would.
Why? What is the crucial difference between the two kinds of experiment? It
is this: inheritance in the drawing experiment is Lamarckian ([. . .] ‘copying-
the-product’). In the origami experiment it is Weismannian ([. . .] ‘copying-the-
instructions’). In the drawing experiment, the phenotype in every generation is
also the genotype – it is what is passed on to the next generation. In the origami
experiment, what passes to the next generation is not the paper phenotype but
a set of instructions for making it. Imperfections in the execution of the instruc-
tion result in imperfect junks (phenotypes) but they are not passed on to future
generations: they are non-memetic.
[. . .] The instructions are self-normalising. The code is error-correcting
(Dawkins 1999: x–xii)
1. Take a square sheet of paper and fold all four corners exactly into the middle.
2. Take the reduced square so formed, and fold one side into the middle.
3. Fold the opposite side into the middle, symmetrically.
4. In the same way, take the rectangle so formed, and fold its two ends into the
middle [. . .] and so on (Dawkins 1999: xi)
to identify some of the steps which the production involves. This will
be because her mind already hosts instructions or memes for those. For
example, she will already know what it means to fold paper. She will have
a concept of ‘corners’, a concept of the ‘centre’, and so on. It is likely that
some of these (possibly themselves memetic) components may indeed be
thought of as discrete units. They only need to be re-associated in a par-
ticular manner to become copies of ‘instructions’, and to establish a more
complex, larger meme, or memeplex. In this sense learning how to fold a
Chinese junk model is exactly analogous to learning a new morphologi-
cal Gestalt by memorising a particular association of pre-defined phone-
mic units or phone-memes, or to learning a phone-meme by associating
pre-defined auditory impressions with pre-defined articulatory gestures.
Now, when the child attempts to fold a junk model herself, working from
memory, her brain will re-organise and arrange the micro-instructions (or
memes) which it has already incorporated into a larger structure. Then,
she/her brain may ‘test’ the resulting macro-instruction (or memeplex)
by getting it expressed. If this achieves the desired result, she may have
another go, thereby strengthening the memory of the memeplex which
she has by now tentatively internalised. After a couple of attempts, she
will be satisfied, and the instructions, the memeplex, or macro-meme,
for making a Chinese junk will be stably represented in her brain. Now
will she be ready to turn to the next child in the sequence and pass them
on to him.
If, on the other hand, her first attempt at making a junk fails, she
will feel that she has arranged the necessary micro-memes in a wrong
way, or employed an incomplete or wrong set of them. Their current
arrangement will therefore not be memorised. Instead, the child will ask
her teacher to demonstrate the process again, perhaps more slowly, or she
will ask for explicit instruction on a step she is unsure about. Thereby,
she will be able to dismiss certain micro-instructions, adduce others,
or possibly even acquire new ones (although this may be harder than
working with known ones). Eventually, she will come to replace the first,
inadequate arrangement of micro-instructions that has formed in her
brain with one that is better at the job. By repeating the process as many
times as necessary, she will be able to restructure her brain until it actually
hosts an adequate copy of the master, or parent-memeplex, sitting in the
brain of her instructor.
Thus, a copy of a complex neuronal structure gets formed in a brain
through a combination of processes which may also reflect Darwinian
principles. From a variety of structures first assumed, or generated, those
which are evaluated as more adequate when their expressions are tested
come to be retained and strengthened. Depending on the degree of their
206 Selfish Sounds and Linguistic Evolution
+
Decreases increases assumes
42 As Henry Plotkin pointed out, following R. C. Lewontin (1970) and D. Campbell (1960),
the brain seems to operate like ‘a “Darwin machine”. That is [. . . its] transformation
in time that occurs through the workings of the processes of learning and intelligence is
the result of evolutionary processes operating within the brain’ (1994: 83). Plotkin’s own
way of schematising the general ‘principles [. . .] that describe the evolutionary process’
(ibid.) is in terms of what he calls a ‘generate-test-regenerate’ (84) heuristic, but indeed
other models such as the ‘Complex Adaptive System’ schema advocated by Gell-Mann
(1992) (see also page 95, above) are equally adequate, and sometimes slightly more
sophisticated.
43 This is also why imperfections in the behavioural expressions of cognitive instructions
won’t normally be transmitted. If, during a demonstration, a child drops the model junk
which it is making, the child that is being instructed won’t interpret this as a step in the
production, and will not copy it either. If I cough while showing my son a dog, he will not
think the animal is called do[kchkch]og. Performance errors are thus never transmitted.
Towards an evolutionary theory 207
often wind up with brain-states that are copies of states realised in other
people’s brains? What we have argued so far might explain how brains
learn and why they preferably learn beneficial rather than harmful things,
but it does not yet seem to explain why their ‘quality space’ should be
biased so as to make them copy from each other. To understand this, we
need to find reasons why brains should be made so as to feel particularly
good when they emulate one another.
46 Note that this question is not the same as asking what instincts may have evolved ‘for’
memetic replication and evolution. Specifically, it does not pre-suppose that the dif-
ferential replication and evolution of memes should necessarily be beneficial to the
genomes that have evolved to make it possible, nor that the ‘phenotypic property of
making memetic evolvability possible’ must have been selected for on the genetic level.
Instead, we are asking whether there may have been instincts, or emotional preferences
for specific brain-states and types of behaviour, that are evolutionarily plausible, that is,
beneficial to the replication of genomes coding for them, and that may have had the effect
of turning humans and their brains into machines for replicating neuronal, structures so
faithfully that they started to evolve along Darwinian lines themselves.
216 Selfish Sounds and Linguistic Evolution
such brain-states should have evolved is highly plausible from the genetic
point-of-view. After all, our co-speciates are also our closest competitors,
and when they are up to something it may clearly pay to be as good
as they are – in whatever it happens to be. Therefore, it will generally
pay to keep an eye on the behaviour of one’s fellows. What it pays to be
particularly watchful of is anything with the characteristics of purpose or
design. After all, any skill or artefact might be turned against us, or might
give our neighbours an edge when it comes to competing for food, or
defending oneself against a predator. Thus, whenever we spot behaviour
that looks structured and systematic, it will pay to try and discover what it
might be for. If indeed it turns out to be useful, it will then pay to become
as good in it as the fellow from whom we have picked it up. An instinct
for keeping-up-with-the-Jones’s makes obvious evolutionary sense.
Consider a possible instinct to conform next. Clearly, it does not
seem to be as easy to confirm through introspection as the instinct for
keeping-up-with-the- Jones’s. Often, we experience the need to conform
as unpleasant and in conflict with our personal intentions and interests.
Yet, the fact that we do feel under pressure to conform at all, even if it
might conflict with other preferences, seems to suggest that it is emo-
tionally relevant to us. In some sense, we do seem to ‘like’ brain-states
whose behavioural expressions will persuade our co-speciates that we are
like them. Such a preference may have evolved in response to an instinct
whose evolutionary plausibility is more or less uncontested. It represents
the core of socio-biological explanations of altruistic behaviour (from
self-sacrifice among social insects to parental care) (Ridley 1996). It is
an instinct for behaving more co-operatively towards kin than towards
non-kin. Since genetic similarity tends to lead to phenotypic similarity,
organisms (of all species) will tend to behave more altruistically towards
co-speciates that look and behave like them than towards such that do
not. Consequently, adapting one’s behaviour so that it looks like that of
others is highly likely to pay. 47
47 The need to conform behaviourally to the co-speciates one is socialised among, may also
be important to humans for a further reason. Ever since their emergence as a species
they have lived in exceptionally large groups characterised by a high degree of social
organisation, the division of labour and the trading of favours. The evolutionary success
of such a life-style depends crucially on being able to distinguish members of one’s
own group from outsiders, so as to avoid dishing out a favour to an individual who will
disappear before paying it back. Therefore, the development of behavioural idiosyncrasies
learnable only by individuals who are around to see them, and thus common only to
group members, may plausibly have become an evolutionarily stable strategy. In fact,
Robin Dunbar (1996) has argued that providing such a type of freely variable behaviour,
may have been the most important factor behind the evolutionary emergence of the
human language capacity.
Towards an evolutionary theory 217
48 All academic staff members who happen to be extraordinarily skilled at dealing with
computer problems will know what I am referring to. The pleasure of being more skilled
than the others soon gives way to the sobering insight that one has become the depart-
ment’s computer handyman. As soon as one realises that all the others get to use their
computers while oneself spends most of one’s time fixing them one finds oneself taking
every chance one gets for imparting some of one’s skills to others.
218 Selfish Sounds and Linguistic Evolution
49 It might be objected that the notion of a strong human imitation instinct makes the coun-
terfactual prediction that humans should attempt to imitate practically every behaviour
they witness, which they don’t. Instead, they are rather selective in what they acquire
through imitation. Therefore, the alleged instincts for imitation cannot be very strong,
Towards an evolutionary theory 219
6.5.10 Summary
We can summarise our theory of how and why some neuronal configura-
tions manage to replicate, and qualify as Dawkinsian ‘memes’. Basically,
the process works like this.
Let a potential ‘meme’ be a neuronal structure with behavioural expres-
sions (these expressions may or may not be artefacts in the narrower
sense). When expressed, its behavioural and/or material products may be
observed by another person. Being human, the mind of the hypothetical
observer is genetically programmed to pay special attention, and be per-
ceptually sensitive to the behaviour of other humans and its results. It is
programmed to feel good when it identifies the properties and purpose of
observed behaviour, and when it feels able to reproduce it, and it is pro-
grammed to feel bad when it perceives the behaviour as being purposeful
yet does not understand it, and/or finds it to be beyond its own capaci-
ties. Thus, when a human comes to perceive the expressions of a meme,
her mind will automatically check if it also incorporates a structure that
could be ‘for’ the observed behaviour or products. This process will not
necessarily be conscious, of course, nor does it need to be regarded as
distinct from perception at all. In fact, higher-level perception itself must
always involve the neuronal activation of structures which are, in a sense,
‘for’ the observed behaviour or artefact.
Saying that some instance of observed behaviour is not only perceived
but ‘recognised’ in the observer’s mind, is the same as saying that an
existing complex of neural configurations in that mind is activated. This
activation will incur positive emotional feedback – indicating to the mind
that all is under control and well, so to speak. Associations among the acti-
vated configurations will be strengthened and stabilised. In such a case,
the observer’s mind already hosts a ‘meme’ for the observed behaviour.
and the whole notion that human minds are genetically determined to act as meme repli-
cation machinery must therefore be wrong.
However, this objection is not really an objection at all. Saying that their instincts
for imitation turn every human brain state with an expression into a potential replicator,
does of course not imply that all of them will manage to replicate. Both brain space
and the energy required for the physical re-organisation of minds must be limited, so
not everything that can be learnt, acquired or imitated will. Were this not the case, Dar-
winian evolution could clearly not occur at all. Selection would not occur, and differences
among replicators could make no difference to the success of their replication. As we
have seen, however, imitation is costly, brain space is naturally limited and therefore no
brain can afford to copy every behaviour it gets informed of. The constraints on what it
will actually come to copy must represent the core of an evolutionary theory of culture,
of course. In section 6.6 we have already discussed a few. Thus, the objection that we do
not actually imitate as much as we might if our instincts for imitation were overpowering
is in fact an argument for a Darwinian approach to cultural evolution.
220 Selfish Sounds and Linguistic Evolution
6.7 Summary
Most of the factors that may constrain the replication and the stability
of memes can be assumed to be constant enough – from their point of
view – to allow sufficient time for adaptive evolution to take place. None
228 Selfish Sounds and Linguistic Evolution
230
Implications for language change 231
replicators was at all plausible, and in what way they might be materi-
ally implemented. The view of human minds as networks of neuronal
cell assemblies turned out to provide a basis for modelling, albeit only
tentatively, the possible material shapes of replicating neuronal con-
stituents for language. It was next shown that they were likely to be long-
lived and fecund enough, and to reproduce with sufficient fidelity to be
capable of historical evolution, and adaptation. Like in other systems
of their kind, the mechanisms by which the historical evolution of lan-
guages was brought about, appeared to involve the creation of varia-
tion though quasi-random ‘copying errors’ or ‘mutations’ and the sub-
sequent automatic selection of better replicating variants over such that
were worse. Taking up a suggestion by Richard Dawkins, we called the
proposed linguistic replicators ‘language memes’. Finally, a model of the
mechanics by which language memes might be copied was developed and
a typology of constraints on their replication deduced. Thus, the basis
for a generalised Darwinian, or evolutionary approach to language was
established.
The approach raises a number of highly complex issues and forces one
to take perspectives which differ considerably from the common sense
attitudes towards language which most humans, including many linguists,
naturally share. Thus, rather than casting them as tools which humans
may build in their minds and which they can use for communication and
cognition, the Darwinian approach suggests that languages are popula-
tions of mental patterns, or ‘memes’, which form within human minds,
direct human behaviour, and thereby bring their own replication about.
Likewise, it suggests that texts and utterances are not merely formal codes
created by humans, which transport meaning from one mind to the next,
but the external expressions of memes, on which the latter depend for
their existence and reproduction. Most disturbingly perhaps, the relation-
ship between languages and humans as conceived of in common sense
appears to be turned on its head. Languages may not primarily be tools
which humans use for communication and cognition and whose proper-
ties can be derived from their purpose. Instead, languages appear to have a
mode of existence in which humans figure not as their owners, designers,
users, or controllers at all, but rather as their ‘hosts’, ‘survival machines’,
or even as elements in their environment, by which their existence and
replication are constrained but not fully determined. In particular, those
aspects of humans to which language (and other) memes are likely to
be sensitive, appear to be properties of human genomes, bodies, brains
and the composition of the memetic population hosted by them, rather
than the properties of ‘conscious human selves’. Thus, there is a sense in
232 Selfish Sounds and Linguistic Evolution
1 While this assumption would be consistent with the well-established observation that the
meanings which linguistic forms may convey in actual utterances are more diverse than
the meanings they are supposed ‘to carry’ and which one finds listed in monolingual dic-
tionaries, it obviously casts doubt on the sense of such structural approaches to semantics
in which linguistic items are supposed to simply ‘have’ meanings.
In relation to this, note that speakers have no status in a replicator based approach
to language, and that this is difficult to reconcile with all semiotic approaches to lan-
guage that define a sign as ‘something which stands to somebody for something in
some respect or capacity’ (Dressler 1985: 281). If linguistic replicators are neuronal
configurations, the associations among them must clearly be mechanical in nature, and
the level on which they exist is obviously inaccessible to any human observer. Since an act
of interpretation is itself merely the activation of a neuronally provided association, one
could even say neuronal associations provide for their own interpretation simply through
existing. Thus, while it may make sense to say that the graphic shape man may stand to a
person for the sound shape [mæn] and vice versa because in that person’s mind a neuronal
234 Selfish Sounds and Linguistic Evolution
Consider larger mental constituents next. Are there any at all? Does
it still make sense, for example, to assume that there is something like
linguistic as opposed to non-linguistic knowledge? If our minds are pop-
ulations of mental replicators, then what exactly is the difference between
linguistic competence and other aspects of cognition? Are they all hope-
lessly mixed in a single bowl of mental replicator soup? This strikes one
as highly unlikely, yet the approach we have been developing suggests
that this might be a plausible way of looking at it. And what about lin-
guistic competences themselves? Are they not inherently structured? Is
there not a difference between lexical knowledge and grammatical com-
petence? There must be, but how is one to account for this from the
meme’s point-of-view? Do established classifications of linguistic com-
petence into phonology, morphology, syntax, semantics and pragmatics
still make any sense at all? Or is cognitive content randomly distributed
within a complex network without any higher-level topology? As weird as
questions like these may sound, they are definitely justified, and we need
to address them.2
configuration ‘for’ man is associated with a neuronal configuration ‘for’ the sound shape
[mæn], it would be clearly absurd to assume that the neuronal configurations themselves
stand for each other to that person as well. Such an interpretation would only work if
human brains hosted homunculi who could observe the relations between individual neu-
ronal constituents. Of course, they do not, and assuming such ‘inner selves’ inevitably
leads to infinite regress – for also the interpretations performed by mind-internal homun-
culi must have a material basis and who, then, is to interpret the mechanical relations that
obtain there? So, unless one is willing to believe in miracles, many semiotic relations that
linguists have assumed to hold among competence constituents become highly question-
able. Who, for example, is supposed to be the interpreter of a semiotic relation between
allophones (acoustic patterns, sensual impressions) and the phonemes they ‘stand for’, if
phonemes are neuronal configurations? (Dressler 1985: 282)
2 Note, first, that it is not a bad thing when a new perspective questions established concepts.
Science is arguably more about that than about providing unquestionable truths (cf. Casti
1989). Thus, an approach which makes apparently established wisdom questionable is
in principle a good thing, even when it appears as a step back, and raises questions
which have presumably long been answered. This, it seems to me, is particularly true in
language sciences which have had a tendency to elaborate interesting intuitions into fully
fledged theoretical paradigms, complete with sophisticated formalisms and all, without
spending sufficient time and effort on questioning basic assumptions. This is equally true
of traditional attempts to describe the grammars of all types of languages in terms of a
Latin based paradigm, of Neogrammarian attempts to chart the histories of individual
languages in terms of covering sound laws, of dependency grammar’s attempts to describe
sentence structure in terms of the physical model of atoms and molecules, of Generative
attempts to write grammars as logical production systems, as well as of functionalist
attempts to explain the properties of languages from the services they do for their users.
Because of an undue impatience, it seems to me, most approaches to language have
tended to paint themselves into corners from which they found it difficult to get out.
Therefore, the fact that the replicator based evolutionary approach to language which
we have been advocating here presently creates more questions than answers and might
eventually necessitate the re-invention of one wheel or the other should, I am convinced,
be regarded as an asset rather than as a drawback.
Implications for language change 235
3 Types of system-internal co-adaptation are well known in linguistics, albeit not under
this name. Instead their results are referred to as ‘typological adequacy’ or as ‘system
adequacy’ (Dressler 1985). In languages that display an SVO syntax, for example, preposi-
tions are found to be more ‘natural’ or more ‘preferred’ than postpositions. Or agglutinat-
ing languages allow for richer and more productive morphologies than inflecting-fusional
languages (see, for instance, Dressler/Ladányi 1998 and 2000.)
8 How to live with feet, if one happens
to be a morph-meme
8.1.1 Introduction
It is generally acknowledged among historical linguists that – roughly
between 900 and 1300 – phonological changes must have taken place
which altered the distribution of English long and short monophthongs
rather drastically.1 The examples in (23) and (24) illustrate the issue.
Take, first, the words in (23a). All of them have diphthongs as their
stressed vowels, which are generally acknowledged to derive historically
from simple short vowels (23c), with long monophthongs as historically
intermediate forms (23b).
(23) Lengthenings
(a) ModE make, acorn, beaver, cloak; child, hound; whale,
bead, coal
(b) LME māken, ākorn, bēver; cı̄ld, hūnd; hwāl, bēd, cōl
(c) LOE/EME makien, akern, befor; cild, hund; hwœl, bed, col
Conversely, the words in (24a) have short vowels which are assumed to
derive from long monophthongs (24b).
(24) Shortenings
(a) ModE kept, dust, fist; errand, southern
(b) OE kēpte, dūst, f ȳst; ǣrende, sūþ erne
1 There are two reasons why I have picked these particular changes to illustrate how the
evolutionary, replicator-based approach developed in the preceding sections can be put to
work. The first is simply that I happen to be more familiar with them than with most other
changes in the history of English phonology, and the second is that my inability to make
full sense of them within any of the established descriptive and explanatory paradigms
which I have tried out has been decisive in making me seriously interested in evolutionary
theory.
240
How morph-memes live with feet 241
With regard to the lengthenings in words such as OE cild > ModE child
or OE hund ModE hound, – which appear to contradict (25b) – it was
noticed that they occurred only before sonorant+voiced stop clusters
whose constituents were articulated at roughly the same place in the
mouth. These clusters were called ‘homorganic’, defined as exceptions
to (25b) and became the basis of a fourth sound law, namely
(26) Homorganic Lengthening (V → [long] {nd, mb, ŋg, ld, lz, rz,
rð , . . .}5
The lengthenings in words like OE hwœl ‘whale’, OE bed ‘bead’ or
OE col ‘coal’, finally, were not regarded as having been ‘brought about’
by a ‘sound law’ at all, but instead through the analogical transfer of
long vowels that had been created by regular Open Syllable Lengthening
in inflected forms such as hwalas, bedes or coles. This view appeared to
be supported by such singular–plural alternations as the one that can be
observed in ModE staff–staves, and – because CVC lengthenings appeared
to be rare (but see below section 8.5.3) – it has been accepted more or
less until today. Thus, all lengthenings and shortenings appeared to be
accounted for as the theory required, that is, in terms of categorical rules
describing sound laws, or in terms of sporadic replacements that were
morphologically rather than phonologically motivated.
Although it is the most widely known and accepted account of the
relevant changes, however, the traditional description has both empirical
and theoretical problems. One is that the four established sound laws
apparently face a disconcertingly high number of exceptions (for example,
we say /kr æk / ‘crack’ where ‘by law’ we ‘ought’ to say /kre k /, /w nd /
‘wind, n.’ where we ‘ought’ to say /wa nd /, and /i:stən / ‘eastern’ where we
‘ought’ to say /estən /). For Open Syllable Lengthening Donka Minkova
(1982) showed that if the rule is used to relate Modern English word
forms to their Old English predecessors, the number of exceptions it
faces practically equals the correspondences which it predicts correctly.
This is particularly awkward since also lengthenings in apparently closed
syllables, as in col ‘coal’, or bed ‘bead’ have subsequently turned out to
be no less frequent,6 so that it is downright puzzling why the former
should be accounted for in terms of a lengthening rule and the latter not.
This is the empirical problem and it is serious. After all, the main value
of Neogrammarian type sound laws is descriptive. They chart regular
and categorical correspondences between sounds of related languages
5 Vowels were lengthened before the clusters nd, mb, ŋg, ld, lz, rz, rð, etc.
6 See Ritt (1997).
244 Selfish Sounds and Linguistic Evolution
8 Vowels were lengthened in open penultimate syllables, if the final syllable was schwa.
9 Speaking more technically, their overall metrical weight.
248 Selfish Sounds and Linguistic Evolution
10 In fact, my decision to do so was motivated to some degree by the fact that the personal
computer which I had acquired to help me with the mechanical aspects of writing came
complete with a spreadsheet and a charting programme. Since I wanted to discover what
the programmes, which I had not even acquired intentionally, could do, I fed them my
linguistic data, and found, somewhat to my surprise that the patterns which this method
revealed were rather striking.
250 Selfish Sounds and Linguistic Evolution
This rule describes how Old English vowels in alleged inputs to quan-
tity changes relate to their Modern English descendants – as far as their
quantity was concerned. It is empirically adequate (but see below). For
example, it appears simply correct to say that ceteris paribus11 one finds
more lengthened vowels among low vowels than among mid vowels, and
more among mid vowels than among high ones, etc. As the following
chart, based on inputs to Open Syllable Lengthening, illustrates (Ritt
1994: 39), the correlation is so obvious that it would be downright stub-
born to deny that the height of a vowel made a difference for its chance
to be lengthened.
11 That is to say, if one factored certain competing factors (for which see Ritt 1994)
out. The principles on which to do this are inherently problematic. On the issue,
see, for example, Lass (1980), Bermudez-Otero/McCully (1997), or Prince/Smolensky
(1993).
How morph-memes live with feet 251
30
(29)
25
Percentage 20
of long reflexes
among items 15
with stable last
syllables 10
0
high mid low
VOWEL HEIGHT
the concept of sound change – or the concepts behind such expressions as ‘(were)
lengthened’ – refers to the following phenomenon: at one time one group of peo-
ple pronounce certain words of their language in one particular way, while other
252 Selfish Sounds and Linguistic Evolution
people at a different time use similar words to convey similar meanings but pro-
nounce them in a different way. A change can be said to have occurred whenever
in a language a certain role is played by one articulatory target at one time, and
by a different target at another time. [. . . T]his view can be broken up to yield the
following more specific interpretations of ‘sound change’. In the first, it stands for
the mere fact that the latter target can be regarded as the functional equivalent and
thus the temporal successor of the former. In the second interpretation, which is
much stronger, ‘sound change’ stands for all the factors that caused the functional
correspondence between the two elements. [. . .]In both readings, [the phrase
‘sound change’ . . .] is a cover term for a large set of interrelated events. (Ritt
1994: 8)
Thus, I used the term ‘sound change’ to stand, basically, for whatever
processes may have brought a diachronic correspondence between dif-
ferent sounds. But, in retrospect this may have been a hedging strategy
more than anything else. Its inherent vagueness consequently tempted,
or practically invited readers to supply their own personal interpretations
of processes behind diachronic sound correspondences. And with some
interpretations the format of the rule I proposed was predictably incom-
patible. For instance, Bermudez-Otero objected to its statistical nature
on grounds of the principle (quoted from Prince and Smolensky 1993:
197–8) that
Linguistic theory cannot be built on ‘laws’ of this sort, because they are too
slippery, because they contend obscurely with partly contradictory counter-‘laws’,
because the consequences of violating them cannot be assessed with any degree
of precision [. . .]
As his own treatment of the changes show, his objection was based
on a misunderstanding. In good generative tradition albeit with a new
optimality-theoretical formalism, Bermudez-Otero attempted to account
for them in terms of an assumed mental production system which takes
pre-change forms as its input and puts out post-change forms. Thinking
in such terms, he seems to have interpreted my formula as a description
of mental processes as well. It was not intended to model such processes
at all, however, nor was it meant to formulate any type of ‘law’ which
to violate could have consequences in any meaningful sense. Instead, it
really just described quantitative relations between successive populations
of (as I would say now, but didn’t say then: ‘memes for’) word forms.12
The second major problem about my account was empirical. As indi-
cated, formula (28) was derived from a corpus of OE words that qualified
as inputs to one of the four Neogrammarian laws for vowel lengthening
and shortening. It described the correlation to their Modern English
12 Of course, when one fails to make it clear enough what one thinks that language changes
really are, one has only oneself to blame for provoking misinterpretations of that kind.
How morph-memes live with feet 253
descendants correctly. However, the way it was formulated (and the way
in which I understood it), clearly implied that it should apply quite gen-
erally to all items in the English lexicon, not just to those which had been
singled out by the Neogrammarians. Unfortunately, it seemed not to.
I had completely overlooked the fact that my formula predicted words
like OE mon and OE god to show up as ModE /me n / and /əυd / rather
than as /m æn / and /ɒd /.13 Although it subsequently turned out that the
number of words of the mon type which do have long-vowel descendants
is actually greater than had hitherto been assumed, the figures were still
not compatible with the predictions inherent to (28).
Of course, that words of the mon-god type seemed to have been mira-
culously immune to the lengthening force which the parameters in (28)
‘ought’ to have exerted on their vowels is reason for worry. In my case, it
made me worry so much that I eventually came to reconsider my whole
approach to language and language change. In a way, this very book
has been triggered by the fact that rule (28) made wrong predictions. It
motivated me to re-think my first and rather naive interpretation of the
formula, and made me think again about the actual mechanics by which
the parameters in it could have affected the quantities of vowels. It was
only when I had trained myself to think of languages as systems of repli-
cators undergoing Darwinian evolution, however, that I realised what my
Quantity Adjustment rule could be expected to predict, and what it could
not. As we shall see, there is a plausible reason why mon, God and so many
other [CVC] monosyllables appeared to be miraculously immune to the
pressures which the factors in (28) ought to have exerted on them. The
pressures were not simply ‘lengthening pressures’. They did not exert ‘a
pressure on vowels to lengthen’. As the Darwinian perspective we have
developed suggests, they must have been selective pressures on the repli-
cation of morph-memes. As we shall see, they were ‘rhythmic’ in origin,
and selected for morphs that replicated well in predominantly trochaic
expressions. Vowel lengthening was just one of the possible ways in which
morph-meme lineages could adapt to them. [CVC] monosyllables had
an additional option which, if they took it, allowed them to maintain their
short vowels.
13 My thanks go to Ricardo Bermudez-Otero who was courageous enough to point this out
to me at International Conference of Historical English Linguistics in 1996.
254 Selfish Sounds and Linguistic Evolution
segmentally short (or ‘light’) feet, and the other way round. When lexical
morph-memes could pursue the second strategy, on the other hand, the
pressure on their own components to adapt was considerably reduced,
and changes of vowel quantity consequently less frequent.
The next sections will show in detail how the explanation works. As
we shall see, it has none of the problems inherent in traditional accounts
(and most of their revised versions). Furthermore, it allows one to inte-
grate the lengthenings in monosyllabic words such as whale or stave,
which have so far been regarded as uncomfortable irregularities or simply
overlooked.
14 The translation of Starkton, which would gloss ‘strongstress’, as ‘full utterance stress’ is
of course interpretative. It is clear from the larger context, however, that Luick could
not have meant lexical stress, as his description of OSL is part of a chapter which deals
exclusively with the history of lexically stressed vowels (‘Sonanten in Tonsilben’) anyway.
256 Selfish Sounds and Linguistic Evolution
While – or maybe even because – this observation and the facts behind it
are relatively easy to interpret some of their rather interesting implications
have tended to go unnoticed.
as, for instance, in cam ‘came’. In minds whose organisation was affected
by exposure to lengthened expressions of /a /, they caused the emergence
of memes for {have}, {make} and {graze} which did not involve links
to the meme for /a / anymore. Instead they linked to the meme for /a /.
Since the expressions of these mutated morph-memes would be similar to
lengthened expressions of their progenitors with links to /a /, they would
incur enough positive feedback to stabilise the memes. Once the mutated
morph-memes had first emerged in the population, a competition among
/a / and /a / for association to the morph-memes for the words in ques-
tion started. Of course neither of the two vowel memes would have been
aware of this competition. The term merely indicates that some of the
have-make-graze type morph-memes in the population now had links to
/a /s and thereby helped to replicate them, while others involved links
to /a /s and helped to replicate those. Thus, the distribution of link-types
within the population of replicator teams ‘for’ Middle English came to
change over time. At some stages, of course, the competition may have
been undecided, and the distribution of morph-meme variants may have
remained stable, with /a /s prevailing in some brains and /a /s in others,
or /a / and /a / working out ways of sharing the association according to
certain (possibly socio-stylistic) regularities.15 Eventually, however, /a :/s
seem to have ousted /a /s from slots associated to the morph-memes for
words like make, graze – and indeed all others in which Open Syllable
Lengthening was implemented. In the case of have, on the other hand,
/a /s were able to maintain their position against /a :/s. How did they man-
age to do so? If we accept Karl Luick’s view, which is essentially sound,
this was because of have’s frequent occurrence ‘in unstressed positions’.
In ‘unstressed positions’, /a / memes would rarely express as [a ]s, because
‘such positions’ favoured short vowel sounds.
What exactly is ‘an unstressed position’, however? And what does it
mean for a morph-meme, if its expressions will frequently occur in one?
To what degree is the average relative prominence of its expressions an
inherent property of the morph-meme, and to what extent is it environ-
mentally conditioned?16 As I shall argue in the following, at least some
of all the parameters conditioning the prominence of a morph-meme’s
expressions are external to the morph-meme itself. Some of them may
express a configuration which, although it will co-operate with morph-
memes in expression, replicates independently of them, namely a meme-
plex for alternating and isochronic metrical ‘feet’.
15 The latter would themselves have been neuronally encoded. Traditional socio-historical
linguistics of the Labovian type might describe them in terms of variable rules.
16 That is, by factors that are not inherent to the morph-meme itself.
How morph-memes live with feet 259
S w
‘σ’
[timing unit]
Figure 8.1 Another look at a meme for feet.
It merely means that such differences will play no role for the linguistic
meme teams through whose expressions they come about. These teams
include no configurations that specifically react to different degrees of syl-
labic prominence, or code for any of them. Imagine, then, that a mutant
meme emerges in that population which does react, specifically, to syl-
lables whose expressions are more prominent than those in its imme-
diate neighbourhood, and which expresses through, say, increasing the
prominence of every other syllable that gets uttered. The question now
is how well this meme will fare in interaction with the rest of the pop-
ulation. The answer, it seems to me, is rather obvious: other linguistic
meme teams will react positively to its expression, because they basically
‘accept’ prominence peaks wherever they occur. The mutant meme for
alternating prominence patterns, on the other hand, will itself react more
positively to utterances that meet its expectations, cause the memetic
configurations that gave rise to them to receive positive environmental
feedback. Thus, it will cause copies of itself to become more frequent
in the population. Therefore, populations of linguistic replicator teams
which include no meme for different degrees of syllabic prominence are
inherently unstable and likely to be invaded by teams which do include
such a meme.
Now, given that languages are highly likely to include memes for dif-
ferent degrees of prominence, the next question is between how many
degrees these memetic configurations are likely to distinguish and for
what kind of patterns they are likely to code. Again, it is possible to
deduce that memes for distinguishing just two degrees of prominence
and for alternating lifts and dips (henceforth [Sw]-memes) should repli-
cate better than memes for more subtle distinctions and more complex
patterns. The reason is that linguistic memes, although they may them-
selves be encoded digitally (that is, in neuronal cell assemblies which may
either fire or not), depend for their replication on being expressed in artic-
ulatory behaviour and sound. Neither of the two involves discrete units,
however, and can therefore convey information only through analogue
signals, which are highly susceptible to distortion. If there are memes for
many different degrees of relative prominence, these memes will there-
fore come to ‘share’ the range of actual acoustic signals that express and
replicate them, as indicated in the following graph.
Now, if two assemblies ‘share a range of prominence levels’, this will
mean that any utterance stretch whose prominence falls within that range,
may trigger the activation of either of the two assemblies. Since both the
stability of an assembly and its propensity to fire are likely to depend
on the frequency with which it gets activated by a particular signal, a
How morph-memes live with feet 261
Memes A B C D E F G H I
Prominence scale - ← → +
Ideal expressions
Real expressions
show, the preference for alternating rhythm (coded for by the foot
meme) may overrule associations between the constituents of polysyl-
labic morph-memes and constituents for prosodic strength. As this sug-
gests, the rhythmic roles which the expressions of a morph-meme play
will always be co-determined by factors that are, for all practical purposes,
independent of the morph-memes themselves. From the point-of-view of
a morph-meme, they will represent parts of its ‘environment’.
How morph-memes live with feet 265
8.2.2 Generalising the case of have: the adaptive value of ‘regular’ open
syllable lengthenings
If one accepts the explanation just given for the evolutionary stability of
short ME /havə /, one will have to use the very same argument to explain
266 Selfish Sounds and Linguistic Evolution
8.3.1 Introduction
If one looks at the Modern English distribution of lengthened non-high22
vowels in di- and trisyllabic words with open first syllables (as in Ritt
1994), one will not fail to observe that, roughly speaking, the chance of
short-vowel variants to outlive and out-replicate their long-vowel com-
petitors was proportional to the overall metrical weight of the unstressed
syllables which followed the vowels in their expressions. Figure 8.4, based
on the analysis in Ritt 1994, illustrates this.
(1) Words like make or crake ‘crack’– in which the stressed vowel was
followed by a syllable that contained at best an optional schwa23 in
its rhyme – display lengthening in more than 90 per cent of all cases,
making make prototypical and crack highly exceptional.
(2) Words like befor ‘beaver’, bodie ‘body’, or weþ er ‘weather’ – whose
second syllable has remained stable and has come to contain a vowel
or syllabic liquid – lengthened much less frequently: only a quarter
of potential inputs show up long in ModE.
(3) Among words whose second syllables have ended up with a VC
rhyme, such as in capon, bonnet or bottom, lengthening has affected
slightly less than 10 per cent of possible victims.
(4) Of those words whose second syllable is even heavier than that, as
in patient or warrant (both VCC), only 5 per cent have survived in a
lengthened variant.
(5) No lengthened variants have survived at all, finally, of words with
two or more syllables following the stressed vowel. In fact, in some
trisyllabic items, long vowel variants were even replaced by short
vowel competitors (as in southern, errand or holiday).
As we shall see, that long-vowel variants replicated better in such
morph-memes where they occurred before single light syllables (and
worse such in morph-memes where they occurred before heavy (or mul-
tiple) ones) offers itself as a very plausible explanation. Once again, the
rhythmic organisation of Middle English utterances is likely to have played
the decisive role.
22 The fact that high vowels usually resisted lengthening is probably due to the ‘inherent’
shortness of their expressions.
23 That is to say a schwa-meme which was not necessarily expressed by a schwa sound.
268 Selfish Sounds and Linguistic Evolution
100 93
P l 80
r e
o n s
b g h 60
a t o
b h r
i e t
l n e
40
i i n 25
t n i
y g n 20
g 9
5
o o
f r 0
pa
ca
be
m
TR
-5
ak
po
av
tie
IS
e/
n/
er
nt
Y
/w
cr
/w
bo
LL
-20
ac
ea
tto
ar
BL
k
ra
th
m
(C
ES
nt
er
(C
V
(C
(C
C@
V
V
V
CV
CV
CV
)
C)
CC
)
)
8.3.2 Another look at the meme-plex for feet: the timing unit
Stress timing languages such as English are characterised by at least a
tendency towards foot isochrony. There is a reason why a meme for such
a tendency should replicate well. Consider that the [Sw]-meme(-plex)
will tend to express, more often than otherwise, in alternating patterns
of more and less prominent syllables. Thereby it will make [Sw]-patterns
statistically more common in actual utterances than patterns like ([SS],
[Sww], [Swww], or even [Swwwww]). This implies that, again statisti-
cally speaking, prominence peaks will tend to occur at temporally regular
intervals. A meme ‘for’ expecting such regular intervals will therefore
have a good chance of establishing itself within the replicator population.
If it does, it will in turn ‘reward’ any meme which causes its expectations
to be met by timing articulation accordingly. Together, the memes will
reinforce one another symbiotically, so to speak, and may combine into a
stable memetic unit for ‘foot-timing’.24 A unit ‘for’ foot timing will express
Of these it is normally said that the duration t(n+1) − t(n) will tend to be
roughly constant. In particular, | góod old | is supposed to last not much
longer than | góod | – at least not as long as one might guess from the
relatively greater number of segments that it contains. Therefore, good in
the foot | góod old | must be pronounced relatively more quickly than good
in the foot | góod |. This implies, of course, that in | góod old | comparably
less time will be available for the articulation of the individual segments
than in | góod |. Since vowels are by their very nature more flexible with
regard to their duration than stops, it is in their articulation that speed
differences will typically manifest themselves most strongly. Thus, the
[υ ]s in feet like | góod old | will tend to be phonetically shorter than the
[υ ]s in feet like | góod |.
25 The same obviously holds true for all items of the make-type.
272 Selfish Sounds and Linguistic Evolution
8.3.5 Summary
Essentially, all vowel lengthenings and shortenings among Middle English
di- and trisyllabic words reflected selection pressures that were exerted
on the phonological structure of Middle English morph-memes by two
rhythmic principles: fixed left stress and foot isochrony. These princi-
ples represented the combined impact of a memeplex for Middle English
feet. The pressures which that memeplex exerted on the population of
morph-memes generally selected against short vowels in auxiliaries like
have, which they frequently caused to be expressed as metrical dips. They
seem to have favoured long vowels in fully lexical morph-memes (which
tended to be expressed as metrical lifts) when these morph-memes were
also expressed in relatively short feet. And they selected for short vowels in
such morph-memes which tended to be expressed in relatively long feet.
Since morph-meme length and foot length correlated, this amounted to
long vowels being favoured in short morph-memes and short vowels in
long morph-memes. Thus, Early Middle English changes of vowel quan-
tity in disyllabic (and as we have already seen: also trisyllabic) words
represent cases of co-adaptation between memes in the meme teams that
Middle English competences represented. Phonological and morphologi-
cal replicators from the lexical section adapted to a memetic configuration
that governed utterance rhythm.
for English feet were not the only factors which had an impact on the
evolutionary stability of long and short vowel memes. Instead, a consid-
erable variety of other, and sometimes conflicting, factors were at play,
and each of them seems to have favoured the replication of short or long
vowels in quite independent ways. These factors included the sonority
of the consonants or clusters which immediately followed the vowels, the
weight – or possibly the phonetic duration – of these consonant clusters,
as well as the height (and therefore the inherent duration) of the vowels
themselves.
The result of their interaction is essentially what my 1994 Quantity
Adjustment Rule captures.26 As its empirical adequacy (for all items
except [CVC]-monosyllables) shows, the pressures exerted by the fac-
tors just listed amount to a rather plausible and unified description/
explanation not only of Open Syllable Lengthening and its exceptions
but, in fact, of all Early Middle English quantity changes that have tradi-
tionally been described in terms of four separate sound laws.27
What is important, however, is that the undeniable variety of factors
involved should not be interpreted as reducing the independent rele-
vance of any single one of them. If the lineage of a morph-meme failed
to respond to a pressure, this does not mean that it was not subjected to
it. It simply means that it did not pay for the morph-meme lineage to
adapt to it, because other pressures ensured that established variants
replicated better than potential competitors.28 Thus, contrary to what the
established sound laws suggest – by attributing distinct sets of changes
26 The timing unit in the English foot-meme may also have been decisive in the competition
between long and short vowels in words like kēpte and dūst, showing up as Mode kept
and dust and traditionally attributed to a law called Pre-Cluster Shortening. There, the
heavy consonantal codas seem to have consumed too much articulation time for long
vowels to survive before them. Although vowel height may have played an additional
role in the competition, the principle was again very much the same as that behind the
‘lengthening’ of make or the ‘shortening’ of southern. In the lengthenings traditionally
called ‘homorganic’ as in cı̄ld > child, finally, the picture is again only seemingly compli-
cated by the fact that their heavy codas would appear to favour short rather than long
vowels. In fact, their homorganic nature allows them to be pronounced almost exactly
as quickly as single consonants, so that the success of long /i / over short /i/ in lexical
representations of child is as little (or, if you will, as much) of a surprise as the success of
/a / over /a / in representations of make.
27 Since Ritt (1994) contains a sufficiently detailed account of the different factors which
seem to have been relevant in bringing the well studied changes in early Middle English
vowel quantity about, no more shall be said about them here.
28 Significantly, recent approaches to modelling speaker competence in the generative tradi-
tion have developed a formalism by which ‘outputs’, that is, theorems in the production
system which grammar, or competence is supposed to represent, are not derived by
means of categorical rules, but through rivalling and violable constraints, which ‘select’
the optimal one from a set of competing candidates. Constraints are supposed to be
universal and individual languages are assumed to differ because they ‘rank’ constraints
differently. The approach is called ‘Optimality Theory’, and while it appears similar to
274 Selfish Sounds and Linguistic Evolution
(such the lengthenings in words of the make type) to distinct single causes
(such as ‘the open-penultimate-syllable environment’), various factors
favouring either short or long-vowel phonemes exerted their selectional
pressures in parallel. That the relative success of long over short vow-
els seems to have also depended on parameters such as their height or
the sonority of their context, does therefore not mean that it did not
equally depend on rhythmically induced pressures, or vice versa. It just
explains why the impacts of each of the pressures seem to have remained
statistical: individual pressures sometimes added up to one another, and
sometimes cancelled one another out. Contrary to some linguists,29 but
in accordance with established practice in many social and biological dis-
ciplines, I find nothing objectionable in this notion, although the issue
naturally merits a separate and more detailed discussion and would seem
to call, in particular, for the application of more sophisticated mathemat-
ical tools than linguists can typically handle. Since all attempts to account
for Early Middle English changes of vowel quantity in terms of categor-
ical rules have proved empirically more or less inadequate, and in the
absence of convincing counter-arguments, I prefer to conceive of these
changes as statistical phenomena, by which individual lexical items were
not affected in a categorical, regular manner, but merely with certain,
specifiable probabilities.30
8.5.1 Introduction
We have argued that lengthenings in words of the make type were adaptive
responses to pressures emerging from the fact that segmentally short
words were – for rhythmical reasons – pronounced relatively long and
thus favoured the replication of long-vowel phonemes over that of short
ones. This has an interesting, and rather obvious, implication, namely that
the approach advocated here, its theoretical basis is still fundamentally different. In par-
ticular, ‘optimality theoretic’ constraints are supposed to be part of Universal Grammar,
and the mechanics by which they select among rivalling outputs therefore internal to
the grammar, that is, the mind as well. Selectional pressures on linguistic replicators
and teams of such, on the other hand, derive to a considerable extent from expression,
that is, the behavioural and textual products of linguistic competence. The fact that
generative theories distinguish categorically between the study of grammar (Chomsky’s
I-language) and discourse (one of Chomsky’s E-languages), plus their decision to focus
more or less exclusively on the former, force them to postulate mind-internal counter-
parts to mind external phenomena with mental effects if they want to incorporate them
into their competence models.
29 E.g. Lass (1980), Prince/Smolensky (1993), or Bermudez-Otero/McCully (1997).
30 Even though I have to accept that their specification remains very much an open problem.
How morph-memes live with feet 275
Read: the probability of vowel lengthening is inversely proportional to the weight of the weak
syllables ( w ) following it within the same word. The evolutionary account of the relation
seems to represent a straightforward interpretation of the Quantity Adjustment rule or
that, conversely, Quantity Adjustment merely formalises the evolutionary, replicator-
based account.
32 I.e. short in terms of the number of segments in the rhymes of their weak syllables.
276 Selfish Sounds and Linguistic Evolution
1914/21) and has never really been questioned since (except in Ritt
1997b). However, it is inadequate. [CVC] monosyllables did lengthen
quite frequently.33 Probably Luick failed to realise their number because
he was simply convinced that there could not be many of them, so he did
not look too hard. After all, it would have been absurd for Luick to search
for evidence of a process that lengthened vowels in closed syllables, when
he defended a sound law in which a necessary condition for vowels to be
lengthened was their occurrence specifically in open ones. Luick’s ‘belief’
in Open Syllable Lengthening may have been additionally strengthened
by the apparent frequency of ‘open syllable lengthenings’ in other Ger-
manic languages.34 These made the law appear plausible also for English,
while they would have made a law for Lengthening in Closed Syllables
highly exceptional. Thus, it came to be the established view that short
vowels in monosyllables with a [CVC] structure simply could not have
been affected by regular quantity changes during the Late Old English
and Early Middle English periods. That the (assumedly) few cases of
[CVC] items where vowel quantity obviously did change came then to
be considered as mere analogical extensions of changes which ‘really’
happened in disyllabic forms, is simply a logical consequence. Thus, the
Oxford English Dictionary still asserts of Modern English whale (< OE
hwœl ) that ‘The present form whale represents oblique forms (hwalas)
etc’ (OED: sv. whale).
However, vowels were not only lengthened more frequently in [CVC]
items than Luick and his successors had believed, but they also lengthened
much less frequently where they should have, if Open Syllable Length-
ening had been a proper Neogrammarian sound change. In particular,
we have seen in figure 8.4 above that words of the types beaver/weather,
capon/bottom or patient/warrant reflect the assumed ‘law’ only in a minor-
ity of items rather than in all (or at least most) of them. Consequently, the
assumption that whale and similar words got their long vowels through
analogical transfer from inflected forms such as /hwa :ləs / is doubly sus-
picious. Quite apart from the unfalsifiability of explanations which make
unconstrained use of analogy, the very forms that Luick proposed as the
bases of transfer are rather improbable.
60
40
20
0
pa
ca (CV
t
m (CV
G (CV
(C wa ient
be eath V)
TR
ak C
po C
od
av er
w VC
V rra /
IS
e/ @
n/ V
/w C)
-20
er
CV nt
Y
(C
cr )
ha
bo C)
/
LL
ac
le
CC
tto
k
A
m
BL
)
ES
fixed left word stress and the foot-meme did make short vowel phonemes
in segmentally short words evolutionary unstable and did favour their
long competitors.
like God or whale than in words like make. Does this mean that the proba-
bilities of vowel lengthening and shortening in polysyllabic items did not
depend on the number of rhyme segments in their weak syllables after
all?
As I shall argue they did. But not straightforwardly. We committed
an error if we thought they should. One should not generalise from an
observed correlation without taking its explanation into account. Con-
sider how we derived the prediction which the [CVC]-lengthenings fail to
bear out. First, we observed that in all cases except [CVC]-monosyllables
the stability of short vowels was proportional to the rhyme-weight of the
weak syllables in the morph-memes with which they are associated. We
also provided an explanation of this correlation. It was that memes for foot
structure, with which memes for morphs co-expressed, had the effect of
shortening vowel sounds in long feet and lengthening them in short ones.
We observed that the segmental length of the feet in which memes for
polysyllabic morphs are expressed will correlate with their own segmental
length. So the causal relation between the rhyme segments in the weak
syllables of a morph-meme, and the meme for the vowel in its stressed
syllable is very indirect. It does not warrant a straightforward law like
∗
(36) p (Vm → [+long]) ≈ k
weight(w )m
36 And when nothing else inhibited the activation of that node. The activation of the
[S]-node may have been inhibited if it had just fired, for example.
37 If they had not done so, they would have been replaced by memes that did.
How morph-memes live with feet 281
the feet in which they are expressed must reflect the absence of weak syl-
lables by being shorter, on average, than the duration of the feet in which
polysyllables are typically expressed. As will be shown, this is probably
not the case.
In order to understand the exceptional status of monosyllables, con-
sider first the following two sentences. They illustrate once more how
the correlation between foot duration and the number of segments in the
weak rhymes of a morpheme comes about.
(37) a. Every syllable has a structure.
b. Every poem has a structure.
They are identical except that (37a) has the trisyllabic syllable where (37b)
has the disyllabic poem. This difference will be reflected in rhythmically
structured utterances of the two sentences, as (38) shows.
would be natural only for rapid speech. Thus, there seems to be a marked
difference between the ways in which polysyllables and monosyllables are
integrated into the rhythmical structures of English utterances. It reflects
the fact that the [S] and [w] nodes in the foot memeplex increase the likeli-
hood of each other’s activation and thus express, preferably, as trochees.
Since lexical monosyllables are in turn likely to activate the [S]-node,
the syllables whose expression comes to follow theirs will frequently co-
express with the [w]-node. The expressions of the final syllables of poly-
syllabic morph-memes, on the other hand, will usually co-express with the
[w] node and thereby increase the probability that the syllable expressed
immediately after them will co-express with the [S]-node. As one could
say in established terms, stressed monosyllables tend to ‘demote’ follow-
ing syllables where stressed polysyllables tend to ‘lift’ them.
Of course, there are many conceivable utterance configurations where
this tendency will be overridden by conflicting factors. Thus, the first syl-
lables in the expressions of full lexical items will tend to ‘resist’ demotion
while ‘unstressed’ prefixes will tend to ‘resist’ lifting, as (43a) and (43b)
illustrate.
(Joshua IV, 4)
38 First lines represent the original, second lines are modified by the author.
284 Selfish Sounds and Linguistic Evolution
(Genesis , 1)
In feet like these, the meme for foot timing would express God more
probably as [od ] than as [od ]. Thereby it would replicate {/od /}
better than {/od /}. In other words, the competition must have been
How morph-memes live with feet 285
a pretty close shave, and this is exactly what the fact that more than
50 per cent of Old English [CVC]-morph-memes actually do have
Modern English descendants with long vowels (e.g. bed ‘bead’, blœd
‘blade’, col ‘coal’, etc.), bears witness to.
That some [CVC]-monosyllables did evolve into long-vowel variants
indicates that they were as much under pressure to adapt to the ways in
which the memeplex for feet affected their expression as all other morph-
memes in the English lexicon. And this is exactly what our hypothesis
suggests of course. Whenever one of them came to be expressed in an
utterance where it figured in a comparably short foot, the phonetic dura-
tion of the sound which expressed its vowel would be extended, and
replicate a long-vowel meme more easily than a short one. Thus, a con-
siderable number of short-vowel phone-memes lost their slots in morph-
memes for [CVC] items, and long competitors took their places. At the
same time, the fact that [CVC] monosyllables ‘lengthened’ less frequently
than [CVCə ]-disyllables, reflects that the memeplex for feet tended to
demote the expressions of syllables that followed them. The syllables
thus demoted became parts of the same feet as the [CVC]-monosyllables
that headed them, which made those feet comparably long in terms of
(rhyme) segments. Then the meme for foot timing would see to it that
[CVC]-monosyllables were expressed relatively quickly. The expressions
of the vowels in them would be short in duration and replicate short vowel
memes better than long ones.
Thus, our theory of the mechanism by which Early Middle English
changes of vowel quantity were brought about seems to be corrobo-
rated, rather than falsified by the history of [CVC]-monosyllables. As
expected, they were not immune to the pressures which the memeplex
for feet exerted on long- and short-vowel phone-memes and their asso-
ciations to memes for English morphs. If one considers that the rhyth-
mical demotion of following syllables cannot have been as likely in the
expression of polysyllabic items as it must have been in those of monosyl-
lables, the evolutionary account that has been suggested here, ‘predicts’
the very distribution of ‘vowel lengthenings and shortenings’ across the
lexicon of Middle English that we actually do observe, namely the one in
figure 8.5 above. It is thus empirically more adequate than most tradi-
tional ones.
8.6 Summary
As we have seen, the evolutionary approach to language which has been
developed in this book not only provides a better explanation of those
EME changes of vowel quantity which established accounts dealt with,
rather unsuccessfully, in terms of categorical rules or various types. It also
286 Selfish Sounds and Linguistic Evolution
and successful replicators. This means that their expressions had to repli-
cate them well. Since foot-memes were resistant to evolutionary change,
all morph-memes which co-expressed with them were under pressure
to adapt to them. Of course all morph-memes have to co-express with
memes for foot structure and timing. In a language such as English, in
which a particular set of them is safely established, they must have prac-
tically amounted to an environmental constant. It is unlikely that their
impact should have been restricted to causing vowel quantity adjustments
in a number of morph-memes.
Of course, the high evolutionary stability of English memes for foot
structure and timing does not predict that all morph-memes should have
been under pressure to actually change. Many would already have been
sufficiently well adapted as they were – such as morph-memes of types
like have, warrant, or man. In some cases other pressures outweighed
the pressure from foot-memes. For instance, vowel height seems to have
selected against length and for shortness. This explains the stability of
short // and /υ / in words like OE sunu ‘son’ or scipu ‘ship’, or the short-
enings in words like OE fȳst ‘fist’ or dūst ‘dust’. In their case, foot-meme
based pressures can still be assumed to have selected for long vowels,
but the impact of vowel height outweighed their evolutionary effects.
Similarly, long vowels seem to have been selected against when they were
followed by voiceless obstruents. Thus, crack (< ME craken), drop (< ME
drope) and fret (ME < freten) have managed to survive with short vow-
els, even though foot-memes would have selected against them. In sum,
the environmental impact which memes for English feet had on English
morph-memes will have led to actual changes in morph-meme lineages
only when additional conditions were met.
Generalising from this, one may say that (a) factors which exert selec-
tional pressures on the memes in a population are unlikely to be involved
in only a single ‘change’, and (b) individual changes are unlikely to reflect
the impact of a single environmental factor only.
This suggests a new way in which the histories of languages might be
told. Traditionally, they have been understood as chronological sequences
of individual changes. Accounts typically focus on the constituents that
undergo a change. Sometimes, changes are seen as causally related to
each other, sometimes it is acknowledged that local changes appear to
‘conspire’ to have common, global effects, but relations among changes
are usually only investigated after the changes themselves are established
as unified events. The approach we have developed here opens another
possibility. Since we have argued that many linguistic changes will repre-
sent meme-to-meme adaptations, we may attempt to tell their story not
from the point-of-view of the meme-lineages that actually change, but
288 Selfish Sounds and Linguistic Evolution
9.1 Introduction
It is more or less a commonplace among historical linguists that many
of the characteristics of Present Day English have somehow followed
from the fact that in Germanic, the progenitor of English, word stress
came to be fixed on the first, leftmost, syllable of the root. One of the
consequences of this fixing is supposed to have been that word final, that
is, the rightmost, syllables first came to be phonetically backgrounded and
reduced, and then historically lost. This development has in turn been
adduced to explain not only the large number of monosyllables in the core
vocabulary of Present Day English, but is additionally supposed to have
furthered the loss of inflectional endings and thus the typological change
of English from an inflecting towards an isolating language. Therefore, it
can be considered indirectly responsible for the fixing of SVO word order
as well, because the latter appears to have been necessitated by the very
loss of morphological case marking, without which syntactic roles such
as subject and object could not be unambiguously indicated anymore.
Although the decisive role which the fixing of Germanic word stress
seems to have played in the evolution of English on almost all levels is
acknowledged by most linguists, however, the question how exactly it
has exerted its influence has not really been addressed. There are many
possible reasons for that, of course, but one of the most important ones
is the methodological difficulty involved in causally relating a single spe-
cific property that a language assumedly acquired at an early historical
stage with the long-term typological development of one of its daughters,
unfolding itself over a period of almost one and a half millennia. The issue
appears almost too big to address, and therefore the role of fixed root ini-
tial stress in the long-term evolution of English has never really made
it beyond a pedagogically convenient myth. However, the evolutionary
framework which we have been developing in this volume may provide a
theoretical basis which is solid enough to justify a new look at the issue
and to investigate the mechanics behind typological conspiracies.
289
290 Selfish Sounds and Linguistic Evolution
1 An interesting paper on the effects which rhythm may exert on other aspects of a language
is Stampe/Donegan 1983.
The Great Trochaic Conspiracy 291
own stability was so great that other memes were more likely to adapt
to it than vice-versa. To the extent that the expression of morph-memes
coincided with the expression of a trochee, a tendency to become more
trochee-like ought to be empirically detectable in the evolution of their
lineages.
Secondly, it suggests that compensatory changes should be expected
whenever independent factors introduced morph-meme variants into the
English meme pool which turned out to be badly adapted to foot-meme
pressures. The reason for this is what we have already observed: the stable
memeplex for English was not the only factor which affected the evolution
of morph-meme lineages. It caused or prevented actual changes always
in combination with, or against other, and sometimes quite independent
pressures. Recall, for instance, that vowel height seems to select strongly
against vowel length: this selectional pressure works in ways which are
clearly unrelated to the mechanics by which foot-memes select for or
against long vowels. Now, it is clearly conceivable that some independent
factor may select against morph-meme variants that are well adapted to
the foot memeplex, and have them ousted by competitors which are not as
good in that respect. What our view of meme–meme adaptation predicts
for such cases is that new variants will come under strong pressure from
foot-memes. Of course, the change which foot-independent factors had
caused is unlikely to be simply reversed by them. If that were an option,
it would not have occurred in the first place. What can be expected,
however, is that foot-memes will strongly select for variants which are
both compatible with the independent pressure, and better adapted to
foot-based pressures at the same time. In other words, we can expect
compensatory changes whenever a change produces morph-memes that
are worse trochees than their ancestors.
So, we expect memes that expressed as trochee–like word forms to have
become increasingly frequent in the history of English, and we expect
compensatory changes to have occurred whenever meme variants were
introduced that were suboptimal in that respect. One way of testing this
prediction, although admittedly a crude one, is to examine a represen-
tative number of well documented sound changes. To corroborate our
view, their outputs need to be better trochees than their inputs. If their
outputs are worse trochees, then we expect them to be accompanied
or closely followed by changes which restored some of their trochaic
qualities. Only if they are not will they speak against our hypothesis.
The following sections report the results of such a test. The predic-
tions we have formulated seem indeed to be borne out. If one takes a
bird’s-eye look, the evolution of English seems indeed to have selected
for memes that expressed and replicated well as trochees, while it seems
292 Selfish Sounds and Linguistic Evolution
to have punished memes that did not by selecting for better adapted
competitors.
who prefer not to get lost in detail, may want to focus on them when
reading the tables in the following sections.3
(47)
TYPE s w # Wo Wg → →W
∗ wurm+i / D 2>1 3>2 3/ >2/
1 1
− /
2 2
∗ flo:ð+u / D 2>1 3>2 3/ >2/
1 1
− /
2 2
∗ wered+u / D 3>2 3>2 4>3 + +
∗ lirn+unγ +u / D 3>2 5>4 6>5 + +
Where the change affected morph-memes for trisyllabic word forms, its
outputs were more similar to trochees – both in terms of syllable structure
as well as in terms of weight. For disyllabic forms this is not true. At least
in terms of weight, however, the outputs were still within the possible
weight range of trochees. Thus, as far as duration is concerned, the change
did increase the number of English word forms that were equivalent to
optimal feet. At least on the whole, it appears to be compatible with
the assumption that memes for word forms evolved to become better at
replicating as trochees.
(48)
TYPE s w # Wo Wg → →W
∗ xæur+i+ðæ / D 3>2 / 5>4/ 1
+ +
2
4 cf. Lass (1994: 98–102), from where also the examples are taken.
296 Selfish Sounds and Linguistic Evolution
(51) σ1 σ2 And σ
O R O R O R
f i nd e ch i l d
and they would weigh 21/2 and 11/2 moræ respectively.5 The effects of the
change are summarised in table (52).
(52)
TYPE s w # Wo Wg → →W
(53)
TYPE s w # Wo Wg → →W
Since this change was part of the set from which our hypothesis was
derived, the evidence it provides does not really count of course.
5 The crude global picture this section tries to convey of developments in the phonological
history of English does not call for a more detailed discussion of such an analysis, but see
Herbert (1986) or Ritt (1994).
6 Since the change was one of the quantity changes discussed in the previous chapter, this
is no surprise.
298 Selfish Sounds and Linguistic Evolution
The overall weight of the affected forms was reduced by a mora, and
they became more similar to trochees in that respect. The effects of the
change are summed up in table (55).
(55)
TYPE s w # Wo Wg → →W
Also this change was one of the ‘quantity adjustments’ from which our
hypothesis was derived. Like pre-cluster shortening, the evidence it pro-
vides does therefore not really count.
The outputs of CVC lengthening are better trochees as far as their weight
is concerned. That it clearly confirms our hypothesis is no surprise, how-
ever, since it was the change from which the hypothesis was derived in
the first place. Strictly speaking it does not count.
(57)
TYPE s w # Wo Wg → →W
At first sight, this change seems to have been neutral, at best, with regard
to the ‘trochaic conspiracy’ we predict. However, its effects have to be
seen in relation to schwa deletion.
As far as its effects on foot structure are concerned, schwa deletion is obvi-
ously a very ambivalent change. On the one hand, there are cases which
are quite in line with the assumption of a ‘trochaic conspiracy’. Originally
trisyllabic items such as ærende > errand would be good examples. But for
other cases this is not true. Items like goode lost their trochaic structure,
while at least remaining within the trochaic weight range. But morph-
memes of the make type both lost their trochaic structure and became
too light on top. They were not proper trochees anymore, although they
had been at first. Thus, at least where it affected disyllabic word forms,
the outputs of schwa deletion were clearly less trochee-like than their
inputs. What should we make of this?
Consider first that linguists widely agree that English schwa dele-
tion represents a very ‘natural’ change. Syllables which are constantly
expressed in weak prosodic positions will generally receive little articula-
tory energy, their acoustic products will be quiet and difficult to perceive.
It is obvious that no phone-meme will replicate well under such condi-
tions. Morph-memes that include phone-memes of this type will be under
strong competition from variants that do not. It is easily conceivable that
the pressures arising from these circumstances may have outweighed the
impact of the otherwise powerful memes for feet.
As we have said, conflicts among selectional pressures are to be
expected. They are not as problematic in an evolutionary approach as
300 Selfish Sounds and Linguistic Evolution
W S W S
This means that the syllables which followed weak adjectives could hardly ever have
been demoted in expression: after all, they were the first syllables of nouns. In order for
trochees to be expressed, the schwas in weak adjectives had to be expressed as well. Thus,
English foot-memes would have selected particularly strongly for their maintenance. It is
therefore no surprise that weak adjectives retained final schwas longer than other morph-
memes.
The Great Trochaic Conspiracy 301
Thus, the outputs of schwa deletion as such do not reflect the selec-
tive impact of English foot-memes, but the concomitant or subsequent
changes they underwent clearly do. As we shall see below, Open Syllable
Lengthenings were not the only changes which increased the metrical
weight of monosyllabic, ‘underweight’ outputs of schwa deletion, and
brought them back into (or kept them within) a ‘healthy’ trochaic weight
range. Quite a number of epentheses and breakings which occurred dur-
ing the Middle English period seem to have had similar effects. The
least mysterious reason why they all seem to have conspired and worked
towards a ‘common’ goal, is the selective pressure exerted by the memes
for English feet.
9.2.11.2 /j/-breaking11
The change inserted epenthetic vowels before /j/ codas. Often, the coda
was deleted in the process. Examples include dæ > dei or clæg > clei. By
re-arranging the structure of the affected rhymes, the change increased
the overall weight of the affected word forms, because the extra-metrical,
or ambisyllabic word final segments came to be fully integrated into the
rhymes of the affected items. The change brought the affected words
within the weight range of typical trochees, and table (60) summarises its
characteristics:
(60)
TYPE s w # Wo Wg → →W
9.2.11.4 w-vocalisation13
In this change, syllable final /w/s were vocalised as in ewe > yeu(e). The
affected segments became part of the nucleus, and the weight of the word
forms was increased, as table (62) shows. The increase in weight also
counteracted the effects of schwa deletion, and prevented the affected
items from becoming lighter than trochees.
(62)
TYPE s w # Wo Wg → →W
12 See Jones 1989: 155. 14 With schwa deletion being factored in.
13 See Jones 1989: 157. 15 With schwa deletion being factored in.
16 See Jones 1989: 159.
The Great Trochaic Conspiracy 303
9.3 Summary
We have now looked at a number of phonological changes which English
word forms underwent in their evolution from Germanic to Late Middle
English. We have seen how the changes altered the foot structures and
the metrical weights of the different morph-meme types they affected. In
a large number of morph-meme lineages pre-change variants and post-
change variants were about equally good at expressing as trochees. There
was hardly a change, however, which did not make better trochees at least
out of some of the morph-memes it affected, and, apart from schwa dele-
tion, there was no change which made any morph-memes worse trochees.
If one asks for each change if it increased the number of morph-memes
that expressed and replicated well as trochees, one sees that more or less
all of them do. Table (65) shows how clear the picture really is.
(65)
Did it increase the number
CHANGE of good trochees?
The message of chart (65) is clear, then: if sound changes affected the
metrical weight of English morph-memes at all, they seem indeed to
have affected it just as we predicted. Sometimes they shortened them,
sometimes they lengthened them, sometimes they changed syllabic roles
which segments played, but whatever they did to morph-memes, they
(practically) always made them better able to replicate when expressed
as trochees. That it paid for them to replicate as trochees was because
the majority of feet in English utterances were trochees, and memes that
co-express well in the majority of utterance patterns will replicate doubly
well. Since the frequency of trochees in English utterances reflects the
evolutionary stability of specific memes for foot structure and timing, all
sound changes that created morph-memes which replicated well when ex-
pressed as trochees can be regarded as adaptations to those foot-memes.
So, English seems indeed to have undergone the sound changes which
we expected. And our expectations were derived from the Darwinian
approach to language which this book has tried to develop and defend.
What does this successful ‘experiment’ imply, then, for the particular per-
spective we have been advocating? First of all it illustrates, it seems to me,
how fruitful the approach is. Of course, generalisations like the one we
have described in this chapter, could be expressed from any other perspec-
tive as well. That English lexical word forms became more trochee-like
over time represents an empirical fact. It is fairly easy to observe. The
view that linguistic constituents are replicators and will evolve accord-
ing to Darwinian principles, however, can also explain it. As we have
seen, the explanation it provides is principled, coherent, explicit and non-
teleological. A Darwinian perspective on language makes it possible to
understand how sound changes can seemingly conspire to increase the
number of trochees in the language without having to invoke mysteri-
ous teleological laws. For example, it does not have to posit trochees as
ideal feet and assume that such an ideal type may somehow have forced
morphemes to become like it. Instead it allows one to deduce a perfectly
nomological, reductionist explanation. It can be derived from the basic
insight that both morph-memes and memetic configurations for English
foot structure are replicators. Being replicators, they exist because they
replicate well as they are under the specific conditions in which they have
to replicate. The rest follows about as automatically as the properties
of biological life-forms follow from the fact that DNA molecules repli-
cate: both morph-memes and memes for rhythmic structures depend for
their replication on being co-expressed. Therefore, they will necessar-
ily come to interact. Each of the memes that share an expression will
require that this expression should have properties that help to replicate
it. Otherwise it would clearly not replicate through the expression. That is
practically a tautology. Since different memes will have different interests
The Great Trochaic Conspiracy 305
The last two chapters have shown how the replicator-based, evolutionary
approach to language which this book has developed can be put to work in
practice. We have seen that it produces plausible accounts of phenomena
with which other approaches have had considerable problems (chapter 8),
and that it allows one to understand the mechanics behind long-term
trends and conspiracies which can otherwise be difficult to talk about
except informally (chapter 9).
This brings my discussion to its close. What I have tried to suggest in
this book is that language can be approached in Darwinian terms, and
that there are many arguments which speak for it. At the beginning a
rough model of ‘language’ and its various manifestations was developed
which was intended to provide a solid basis for describing and possi-
bly explaining linguistic changes. Then, it was argued that such changes
are best understood as processes in which their constituent properties
are replaced by new variants as they are transmitted from speaker to
speaker and from generation to generation. It was concluded that lan-
guages represented complex, replicating and adaptive systems, and were
in that sense comparable to other such systems, particularly to biologi-
cal species. Next, the essentials of Darwinian evolutionary theory were
discussed, in order to illustrate the concepts and explanatory strategies
which the study of replicating system requires. A long part was then
dedicated to substantiating the assumption that languages could indeed,
and not only metaphorically, be regarded as Darwinian systems. General
properties of complex adaptive systems were described, and the specific
ways in which languages have these properties discussed. Since the inter-
subjectively verifiable existence of stable replicating patterns, which copy
both faithfully and fecundly was shown to be an essential component
of Darwinian systems, it was attempted to show that linguistic repli-
cators were indeed conceivable. Ways were proposed concerning how
it could be determined whether constituents of linguistic systems actu-
ally qualify as replicators, and how such linguistic replicators might be
307
308 Selfish Sounds and Linguistic Evolution
deserve. What then are the major advantages of the Darwinian approach
to language and language change which this book has advocated?
First, it seems to me, it integrates the historical perspective on lan-
guages more seamlessly with what we know both about languages in other
respects, and about the environments in which languages exist. For exam-
ple, it allows one to relate the histories of language systems quite naturally
with the histories of their speakers, without necessitating shifts of perspec-
tive. Thus, consider that we have told the story of Early English vowel
quantities from the perspective of competence constituents and without
taking into account the social details which we know for sure that they
must have been involved in the processes through which ‘English’ has
been passed down from generation to generation of speakers. We did
not talk much about differences in the prestige which the rivalling vowel
quantities must certainly have had in the various subgroups which have
made up the community of English speakers during the last centuries.
However, this does not necessarily mean that we have left out an essen-
tial component of the story which we ought to have been telling. Instead,
our perspective may explain why, in spite of the undeniable fact that pop-
ulations of linguistic replicators will always be under many and diverse
socially based pressures, we can expect that longer-term evolutionary
developments may fail to reflect them. Instead, they will reflect pressures
which different replicators within a system exert on one another. The
reason for this is, basically, that a successful replicator will be much more
long-lived than particular social constellations and therefore be able to
exert its pressures much longer than the latter. This accounts beautifully,
it seems to me, for the puzzling fact that much which is apparently mean-
ingful can be said about the histories of languages even if one takes only
their systemic properties into account even though one knows at the same
time that languages serve social purposes and must therefore be under
pressure to adapt to them. At the same time, a Darwinian, replicator-
based view of language does not rule out that socially based pressures
will play their part as well. Quite the contrary; since language replication
strongly depends on instincts to conform and to express power relations
within groups, socially based pressures will be much more powerful than
most others in the short run. By strongly selecting for specific replica-
tors, to which others will then have to adapt, they may play crucial roles
in defining the course which the longer term evolution of a particular
replicator team will take.
Thus, more generally speaking, our approach defines a perspective
from which there is no essential, qualitative difference between system-
internal and system-external factors in linguistic evolution: namely the
perspective of linguistic replicators themselves. From their point-of-view,
310 Selfish Sounds and Linguistic Evolution
human hosts, but the feedback effects their expressions have on their own
replication.
One of the main reasons why a replicator based, evolutionary approach
to language can provide such a unified perspective on things linguistic is
that many concepts, which have traditionally represented primitives of
most linguistic theories that I know of, have no fundamental status in it.
This applies first and foremost to the concept of the ‘speaker’. As we have
argued it is ambivalent and normally ill-defined. Therefore it has usually
created more problems than it can allegedly solve. Speakers seem to figure
sometimes as mere bodies with a language organ which defines universal
constraints on possible competence structures, and sometimes as free-
willed and creative owners and users of language, who can control what
they say, think and learn. From the point of view of linguistic replicators,
on the other hand, no such categorical distinction is warranted. Instead,
the mind–bodies in which they live represent a single type of environment
with some features which they will perceive as constant and others, which
may appear chaotic and changeable from their point of view, and about
which they may only ‘hypothesise’ in statistical, probabilistic terms.
While redefining our understanding of the speaker–language relation,
a replicator based approach can explain why linguists have usually had
such difficulties when they tried to pin down apparently straightforward
concepts such as ‘language’, ‘dialect’, ‘variety’, ‘style’ and so on (see
Mazzon 2000). It suggests, first of all, that ‘languages’ must be emergent,
higher-level entities which derive from the complex interactions between
the true units in linguistic evolution, namely the replicators themselves.
This clearly explains why they should be impossible to delimit, except in
fuzzy terms. Like all complex replicator systems, languages are dynamical
and open. They may be quite stable and highly organised populations
of smaller constituents, but they are populations nevertheless, in which
members may come and go without the whole falling apart. Thus, instead
of asking what a language (variety/dialect/etc.) is, our approach suggests
that we should better ask how replicators come to self-organise into teams
and what properties we may expect such teams to display. In this respect,
our approach provides an explanatory perspective on concepts which,
traditionally, have been both taken for granted and regarded as infinitely
puzzling at the same time.
Another advantage of the approach which we have been advocating
is its strong commitment to materialism and to psychologically realistic
models of language. Thus, it admits no concepts without interpretation in
terms of constituents of Popper’s World One. Although it cannot change
the fact that the material basis of cognition and knowledge is still poorly
understood, it takes an emphatically monist stance on the mind–body
312 Selfish Sounds and Linguistic Evolution
problem, and serves as a solid basis for developing models of language and
cognition which can – at least in principle – be held against and modified
in accordance with findings made in psychology and neurobiology.
Finally, the compatibility of an evolutionary approach to language with
similar paradigms applied in the study of other and quite diverse empiri-
cal domains, might hopefully facilitate the trade of concepts and method-
ologies between disciplines and help, in particular, to develop models of
language and its evolution which lend themselves more easily to quantifi-
cation than established linguistic theories have tended to do.
Of course, there are many things that could still be said, and this book
may have raised more questions than it has answered. Much though I
would have liked to do so, I could not address them all. I am also aware
that there is a lot of empirical work waiting to be done. Many linguistic
changes in many different languages have been described and accounted
for, sometimes more and sometimes less successfully, by the historical
linguistic community. It remains to be seen if a Darwinian view will shed
any light on them, and if it allows even only to describe them properly.
As I have already pointed out, my hope must therefore be that the views
which have been presented here may inspire colleagues to pursue some of
their implications, to point out the errors that I have definitely committed
and to take the project further.
As Daniel C. Dennett put it in a different context,
The ideas expressed in this book are just the beginning. This has been an intro-
duction to [how] Darwinian thinking [might be applied to historical linguistics],
sacrificing details again and again to provide a better understanding of the overall
shape of [the . . .] idea. But as Mies van der Rohe said, God is in the details. [I am
fully aware of that, and I know that it won’t be necessary for me to . . .]urge cau-
tion alongside the enthusiasm I hope I have kindled in you. I have learned from
my own experience how easy it is to concoct remarkably persuasive explanations
that evaporate on closer inspection. The truly dangerous aspect of Darwin’s idea
is its seductiveness. Second-rate versions of the fundamental ideas continue to
bedevil us, so we must keep a close watch, correcting each other as we go. The
only way of avoiding the mistakes is to learn from the mistakes we have already
made. (1995: 521)
References
313
314 References
Cziko, Gary 1995, Without miracles: universal selection theory and the second
Darwinian revolution, Cambridge, Mass.: MIT Press.
Dahlbom, Bo (ed.) 1993, Dennett and his critics: demystifying mind, Oxford:
Blackwell.
Dalton-Puffer, Christiane and Nikolaus Ritt (eds.) 2000, Words: structure, mean-
ing, function, Berlin: Mouton.
Darwin, Charles 1859, The origin of species by means of natural selection, London:
Murray.
Dawkins, Richard 1982, The extended phenotype, Oxford University Press.
1986, The blind watchmaker, Harmondsworth: Penguin.
1989, The selfish gene, 2nd edition, Oxford University Press.
1993, Viruses of the mind, in Dalhbom (ed.) 1993, pp. 13–27.
1995, River out of Eden, New York: Harper Collins.
1996, Climbing Mount Improbable, Harmondsworth: Penguin.
1999, Foreword in Blackmore 1999, pp. vii–xvii.
Deacon, Terrence W. 1997, The symbolic species: the co-evolution of language and
the brain, New York: Norton.
Dennett, Daniel C. 1990a, ‘The interpretation of texts, people and other arte-
facts’, Philosophy and Phenomenological Research 50, Supplement: 177–94.
1990b, ‘Memes and the exploitation of imagination’, The Journal of Aesthetics
and Art Criticism 48: 126–35.
1993, Consciousness explained, Harmondsworth: Penguin.
1995, Darwin’s dangerous idea: evolution and the meanings of life, New York:
Simon and Schuster.
1999a, ‘Foreword’, in Aunger (ed.) 2001, pp. vii–ix.
1999b, ‘The evolution of culture’ [Charles Simonyi Lecture, Oxford University,
17 February 1999], available online (April 2003): <http://www.edge.org/
documents/archive/edge52.html>.
Diamond, Jared 1998, Why sex is fun: the evolution of human sexuality, New York:
Basic Books.
Donegan, Patricia Jane 1979, On the natural phonology of vowels, Ann Arbor:
Karoma.
Dressler, Wolfgang U. 1985, Morphonology: the dynamics of derivation, Ann Arbor:
Karoma Press.
1999, ‘Why collapse morphological concepts? Open peer commentary on
Clahsen 1999’, Behavioural and Brains Sciences 22: 1021–2.
Dressler, Wolfgang U. and Mária Ladányi 1998, ‘On grammatical productivity
of word formation rules’, Wiener Linguistische Gazette 62/63: 29–55.
2000, ‘On contrastive word formation: German and Hungarian denominal
adjective formation’, in Dalton-Puffer and Ritt (eds.) 2000, pp. 59–75.
Dunbar, Robin 1996, Grooming, gossip and the evolution of language, London:
Faber and Faber.
Dunbar, Robin, Knight, Chris and Camilla Power (eds.) 1999, The evolution of
culture, Edinburgh University Press.
Durand, Jacques (ed.) 1986, Dependency and non-linear phonology, London:
Blackwell.
Dziubalska-Kolaczyk, Katarzyna 1995, Phonology without the syllable: a study in
the natural framework, Poznan: Motivex.
316 References
1996, ‘Natural Phonology without the syllable’, in Hurch and Rhodes (eds.)
1996, pp. 53–72.
Earman, John (ed.) 1992, Inference, explanation, and other frustrations: essays in the
philosophy of science, Stanford: University of California Press.
Eaton, Roger, Fischer, Olga, Koopmann, Willem and Friderike van der Leeke
(eds.) 1987, Explanation and linguistic change, Amsterdam: Benjamins.
Eco, Umberto 1989, Foucault’s Pendulum (translated by William Weaver), New
York: Harcourt Brace.
Edelmann, Gerald M. 1987, Neural Darwinism: the theory of neural group selection,
New York: Basic Books.
1989, The remembered present, New York: Basic Books.
Eldrege, Niles 1995, Reinventing Darwin: the great evolutionary debate, London:
Weidenfeld and Nicolson.
Fill, Alwin 1993, Ökolinguistik. Eine Einführung, Tübingen: Narr.
Fisiak, Jacek (ed.) 1997, Studies in Middle English linguistics, Berlin: Mouton.
Fodor, Jerry A. 1989, ‘Making mind matter more’, Philosophical Topics 17: 59–80.
Gell-Mann, Murray 1992, ‘Complexity and complex adaptive systems’, in
Hawkins and Gell-Mann (eds.) 1992, pp. 3–18.
1994, The quark and the jaguar, New York: Freeman.
1995, Complex adaptive systems, in Morowitz and Singer (eds.) 1995, pp. 11–25.
Gessner, Conrad 1555, Mithridates, Zürich: Christoph Frosch.
Giegerich, Heinz J. 1985, Metrical phonology and phonological structure, Cambridge
University Press.
Gold, Ian and Daniel Stoljar 1999, ‘A neuron doctrine in the philosophy of
neuroscience’, Behavioural and Brain Sciences 22: 809–30.
Goldsmith, John A. 1976, Autosegmental phonology, Bloomington: Indiana
University Linguistics Club.
Gould, Stephen Jay 1982, ‘Darwinism and the expansion of evolutionary theory’,
Science 216: 380–7.
1983, The panda’s thumb: more reflections on natural history, Harmondsworth:
Penguin.
1989, Wonderful life: the Burgess Shale and the nature of history, New York: Norton
1996a, Dinosaur in a haystack: reflections in natural history, London: Jonathan
Cape.
1996b, Life’s grandeur: the spread of excellence from Plato to Darwin, London:
Jonathan Cape.
Harris, John 1994, English sound structure, London: Blackwell.
Hartmann, Peter 1963, Theorie der Sprachwissenschaft, Assen: Van Gorcum.
Haspelmath, Martin 1999, ‘Optimality and diachronic adaptation’, Zeitschrift für
Sprachwissenschaft 18: 180–206.
Hawkins, John A. and Murray Gell-Mann (eds.) 1992, The evolution of human
languages, New York: Addison-Wesley.
Hofstadter, Douglas R. 1995, ‘The architecture of Jumbo’, in Hofstadter et al.
(eds.) 1995, pp. 87–126.
1979, Gödel, Escher, Bach: an eternal golden braid, New York: Basic Books.
Hofstadter, Douglas R. and the Fluid Analogies Research Group 1995, Fluid
concepts and creative analogies, New York: Basic Books.
References 317
Lakoff, George 1987, Women, fire and dangerous things: what categories reveal about
the mind, Chicago University Press.
Langacker, Ronald W. 1987, Foundations of cognitive grammar, vol. I. Theoretical
prerequisites, Stanford University Press.
Lass, Roger 1994, Old English, Cambridge University Press.
1976, English phonology and phonological theory, Cambridge University Press.
1996, ‘Of emes and memes: on the trail of the wild replicator’, VIEWS 5: 3–11.
1997, Historical linguistics and language change, Cambridge University Press.
1980, On explaining language change, Cambridge University Press.
1987a, ‘Language, speakers, history and drift’, in Eaton, Fischer, Koopmann
and van der Leeke (eds.) 1987, pp. 151–76.
1987b, ‘On sh∗ tting the door in Early Modern English: a reply to Professor
Samuels’, in Eaton, Fischer, Koopmann and van der Leeke (eds.) 1987,
pp. 251–6.
1990, ‘How to do things with junk: expatiation in language evolution’, Journal
of Linguistics 26: 79–102.
1997, Historical linguistics and language change, Cambridge University Press.
Lehmann, Winfred P. and Yakov Malkiel (eds.) 1968, Directions for historical lin-
guistics, Austin: University of Texas Press.
Lewontin, Richard C. 1970, ‘The units of selection’, Annual Review of Ecology
and Systematics 1: 1–18.
1982, Human diversity, New York: Scientific American Library.
Liberman, Anatoly 1992, ‘A bird’s-eye view of open syllable lengthening in
English and in the other Germanic languages’, NOWELE 20: 67–87.
Lightfoot, David 1991, How to set parameters, Cambridge, Mass.: MIT Press.
1996, The development of language: acquisition, change, and evolution, Malden,
Mass.: Blackwell.
Lorenz, Konrad 1973, Vergleichende Verhaltensforschung: Grundlagen der Ethologie,
Berlin: Springer.
Luick, Karl 1898, ‘Beiträge zur englischen Grammatik III: Die Quan-
titätsveränderungen im Laufe der englischen Sprachentwicklung’, Anglia 20:
335–62.
1914/21, Historische Grammatik der englischen Sprache, Stuttgart: Tauchnitz.
Lumsden, Charles J. and Edward O. Wilson 1981, Genes, mind and culture,
Cambridge, Mass.: Harvard University Press.
Lynch, Aaron 1996, Thought contagion. How belief spreads through society: the new
science of memes, New York: Basic Books.
Margulis, Lynn 1981, Symbiosis in cell evolution, San Francisco: Freeman.
1989, Did Darwin get it right? Essays on games, sex and evolution, New York:
Chapman and Hall.
1996, ‘Evolution – natural and artificial’, in Boden (ed.) 1996, pp. 173–8.
Mazzon, Gabriella 1997, ‘The study of language varieties in diachrony and
synchrony, or: on methodological cross-fertilization’, in Fischer and Ritt
(eds.), available online (April 2003): <http://www.univie.ac.at/Anglistik/hoe/
pmazzon.htm>.
2000, ‘Describing language variation in synchrony and diachrony: some
methodological considerations’, VIEWS 9: 82–104.
References 319
2000, ‘The spread of Scandinavian 3rd Psn. Pl. Pronouns in English: opti-
mization, adaptation, and evolutionary stability’, in Kastovsky and Mettinger
(eds.) 2000, pp. 279–305.
Rose, N. (1998). ‘Controversies in meme theory’, Journal of Memetics 2,
available online (April 2003): <http://www.cpm.mmu.ac.uk/jom-emit/1998/
vol2/rose n.html>.
Rose, Steven 1993, The making of memory, 2nd edition, New York: Doubleday.
Rumelhart, David L., McCleland, James L. and the PDP research group (eds.)
1986, Parallel distributed processing: explorations in the microstructure of cognition,
Cambridge, Mass.: MIT Press.
Russell, Bertrand 1961, History of western philosophy, 2nd edition, London:
Routledge.
Samuels, Michael L. 1987a, ‘The status of the functional approach’, in Eaton,
Fischer, Koopmann and van der Leeke (eds.) 1987, pp. 239–50.
1987b, ‘A brief rejoinder to Professor Lass’, in Eaton, Fischer, Koopmann and
van der Leeke (eds.) 1987, pp. 257–9.
Saussure, Ferdinand de 1974, Course in general linguistics (transl. by C. Baltaxe),
London: Fontana.
Schendl, Herbert 1996, ‘Who does the copying? Some thoughts on N. Ritt’s
“Darwinising historical linguistics” ’, VIEWS 5: 47–50.
Schleicher, August 1863, Die Darwinische Theorie und die Sprachwissenschaft,
Weimar: Böhlau.
Simpson, John and Edmund Weiner (eds.) 1989, The Oxford English dictionary,
2nd edition, Oxford: Clarendon Press.
Sinclair, John McHardy 1992, ‘The automatic analysis of corpora’, in Svartvik
(ed.) 1992, pp. 379–97.
Snyder, Solomon H. 1986, Drugs and the brain, New York: Scientific American
Library.
Spitzer, Manfred 1996, Geist im Netz. Modelle für Lernen, Denken und Handeln,
Darmstadt: Wissenschaftliche Buchgesellschaft.
Stampe, David and Patricia Jane Donegan 1983, ‘Rhythm and the holistic orga-
nization of language structure’, in Richardson and Cluterman (eds.) 1983,
pp. 337–49.
Stevick, Robert D. 1963, ‘The biological model and historical linguistics’,
Language 39, 159–69.
Svartvik, Jan (ed.) 1992, Directions in corpus linguistics, Berlin: Mouton de Gruyter.
Trudgill, Peter 1974, The social differentiation of English in Norwich, Cambridge
University Press.
Van Rootselaar, B. and J. E. Staal (eds.) 1968, Logic, Methodology and Philosophy of
Science III. Proceedings of the third International Congress for Logic, Methodology
and Philosophy of Science, Amsterdam, 1967, Amsterdam: North-Holland.
Vennemann, Theo 1972, ‘On the theory of syllabic phonology’, Linguistische
Berichte 18: 1–18.
1986, Neuere Entwicklungen in der Phonologie, Berlin: Mouton de Gruyter.
1988, Preference laws for syllable structure and the explanation of sound change,
Berlin: Mouton de Gruyter.
Vincent, Nigel 1986, ‘Constituency and syllable structure’, in Durand (ed.) 1986,
pp. 305–19.
322 References
Vrba, Elizabeth S. (ed.) 1985, Species and speciation. Transvaal Museum Mono-
graphs 4, Pretoria: Transvaal Museum.
Waldrop, Mitchell 1993, Complexity: the emerging science at the edge of order and
chaos, London: Viking.
Weinreich, Uriel, Labov, William and Marvin Herzog 1968, ‘Empirical founda-
tions for a theory of language change’, in Lehmann and Malkiel (eds.) 1968,
pp. 95–195.
Zeki, Semir 1993, A vision of the brain, London: Blackwell.
Index
323
324 Index
mora 292, 294, 295, 296, 297, 298, phone memes 133–134, 170, 171, 172,
301–303 183, 185, 205, 257, 299
definition 292 knowledge vs. expression 172
morph-memes 170, 172, 173, 174, 180, shape of 171
183, 185, 223, 224, 236, 240, 241, phoneme clusters 134
242, 248, 253, 254, 255, 258, 263, phoneme sequences 134, 135, 138, 139,
264, 265, 266, 267, 269, 270, 271, 140, 141, 143, 150, 153
272, 273, 275, 277, 278, 279, 280, phonemes
282, 283, 284, 285, 286, 287, 288, as digitisers 203
290, 291, 293, 294, 295, 299, 300, mental shape of 148–149
303, 304, 305, 306 phonotactic preferences 292
shape of 174 phonotactics 140, 142, 145, 176, 177,
Mufwene, Salikoko 58 178, 257
Murray, Robert W. 93, 141 Pinker, Steven 7, 100, 136, 137, 169, 222
mutation 65, 68, 71, 74, 75, 78, 84, 86, Plato 33
91, 92, 93, 199, 208, 209, 226, 231, Plotkin, Henry 98, 161, 206, 211, 213,
241 214
Popper, Charles 5, 7, 26, 33, 159, 311
natural selection see selection population of competences 46, 47, 98,
Neogrammarian type correspondence 104
rules 242 Pre-cluster Shortening 242
Neogrammarianism 245 prediction
Neogrammarians 40, 253 in evolutionary theory 83–84
neuronal assemblies 93, 135, 161, predictive laws 83
163, 165, 167, 168, 172, 180, preferences 52, 140, 177, 190, 192, 193,
223, 224, 231, 260, 261, 262, 194, 210, 213, 214, 215, 216, 223,
263, 264 227
neuronal realism 125 Prince, Alan 245, 250, 252, 263, 274
neurones 92, 160, 161 progress in language change 101–103
nomological explanation 304 prominence node 181
nucleus 141, 142, 175, 176, 177, 178, prominence peak 143, 181
302 property replication 48, 56, 121
prototypes
OE Shortening before Consonant conceptual prototypes 136
Clusters 296, 303 proverbs 130
onset cluster memes psychological reality of competence
shape of 176 models 107
onset maximal syllabication 293
Open Syllable Lengthening 242, 243, 244, Quantity Adjustment 249, 250, 251, 253,
247, 248, 249, 250, 251, 255, 258, 273, 275, 296
262, 268, 273, 276, 298–299, 300, Quirk, Randolph 32
303
Minkova’s version 247 raising of /e / 39–47
optimality in evolutionary theory 84 random mutation x, 65, 69, 70, 78
Orton, Harold 43, 44 reciprocal altruism 217
Oxford English Dictionary 43, 45, 276, reductionism 80–81
277 reductionist explanation 304
regular correspondences 40, 245
Paley, William 63 replicating systems 48–53, 54, 55, 56, 57,
Paterson, Hugh 82 58, 59, 60, 103, 307
Paul, Hermann 22, 24, 35, 36, 55, 56, theory of 60
162 replicator alliances 71, 74, 75, 134, 139,
Penrose, Roger 159 144, 145, 153, 233, 263
phenotypic expression 97 replicator perspective vs. product
Phillips, Betty 246 perspective 206–208
328 Index