How to Do Things with Junk: Exaptation in Language EvolutionAuthor(s): Roger LassSource: Journal of Linguistics, Vol. 26, No. 1 (Mar., 1990), pp. 79-102Published by: Cambridge University PressStable URL: http://www.jstor.org/stable/4176037 .Accessed: 16/05/2011 00:48

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=cup. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.

Cambridge University Press is collaborating with JSTOR to digitize, preserve and extend access to Journal ofLinguistics.

http://www.jstor.org

J. Linguistics 26 (1990), 79-102. Printed in Great Britain

How to do things with junk: exaptation in language evolution1

ROGER LASS

Department of Linguistics, University of Cape Town

(Received I5 May I989; revised i3 September I989)

Eine ... wichtige Eigenschaft aller Lebewesen, die wir bei der Kenntnis der Spielregeln des evolutiven Geschehens ohne weiteres verstehen, ist die groBe 'Konservativitat' ihrer Strukturen. Durch eine Veranderung der Lebensweise, besonders wenn sie Anpassungen an einen neuen Lebensraum erfordert, konnen alte Strukturmerkmale sinnlos werden. Konrad Lorenz (1978: 23)

i. EXAPTATION

One of the less rewarding of our common interdisciplinary pursuits is lifting theoretical concepts from subjects not our own, and using them in contexts very distant from those they were intended for. Such borrowings often turn from theoretical claims into sloppy metaphors, leading to varieties of 'vulgar X-ism', the result of overenthusiastic appropriation with insufficient sense of the subtlety or precise applicability of the originals. Spencer's 'Social Darwinism', vulgar-Freudian or vulgar-Marxist literary analysis and sociology are nice examples. Linguistics, being less unique than linguists often think, is no exception: Praguian and neo-Praguian functionalism may be a kind of vulgar Darwinism, extending notions of 'adaptation' or 'selective pressure' to the inappropriate domain of language systems (see Lass, ig80a). But every once in a while such transfers seem to work, like Darwin's borrowings from late eighteenth-century Scottish economic theory; if not always through direct applicability, then by focusing on new ways of

[i] Oral presentations of this material were given at the universities of Stellenbosch, Cambridge, Manchester, Edinburgh and Durham. I have profited by comments from Rudie Botha, Melinda Sinclair, Gabriel Dover, Bob Coleman, Nigel Vincent, Richard Hogg, Jim Hurford, and Charles Jones. A preliminary version was published in Stellenbosch Papers in Linguistics (I988), unfortunately before the oral deliveries, which helped to straighten out some mistakes and unclarities. Nigel Vincent and two nameless referees for JL read a later draft and commented helpfully on matters of substance and style. I am also grateful to Roy Pfeiffer for advice on matters Netherlandic, and in an impersonal though no smaller way to the writings of Stephen Jay Gould, who though a palaeontologist and not a linguist is pre-eminently a historian, with a fine eye for the kinds of problems and ideas that make historiography worthwhile. One note on usage: the term 'evolution' here never has the vulgar progressivist sense of 'directional change with increasing "fitness"''; the evolution of a system is simply the story of its change over time, normally the product of variation and differential selection of variants. It is in this sense only that any biological parallels are to be taken (and see ? 4 below).

79

ROGER LASS

interpreting old data, or providing a basis for linking disparate phenomena as instances of a new (putative) natural kind.

The term 'exaptation' comes from evolutionary biology; it was coined by Stephen Jay Gould and Elisabeth Vrba (I982) as, in their title, 'a missing term in the science of form'. In a popular summary, Gould writes:

We wish to restrict the term adaptation only to those structures that evolved for their current utility; those useful structures that arose for other reasons, or for no conventional reasons at all, and were then fortuitously available for other changes, we call exaptations. New and important genes that evolved from a repeated copy of an ancestral gene are partial exaptations, for their new usage cannot be the reason for the original duplication (I983: 171).

The reference here is to the presence in the cells of many organisms of large amounts of redundant or 'junk' DNA, in the form of duplicate genes (often hundreds of copies: Gould and Vrba note that about a quarter of the total genetic material of fruit-flies and humans exists in such a form). They argue (I982: IO f.) that this surplus DNA is of immense evolutionary importance: it serves as a locus for change which - because the genes in question are not at the moment being expressed - is neutral with respect to actual form or function; it occurs as a kind of background process, while the original DNA goes about its business of coding for currently relevant structures and functions. This mutated genetic material however CAN at some point be expressed - it is as it were covertly evolving. Most important from a theore.tical point of view however is the notion that - unless we are prepared to invoke 'backwards causation' - it could not have evolved 'for the purpose of' providing such a reservoir for future genetic change. Its use for such things is in Gould and Vrba's terms an exaptation.

The concept may perhaps be clarified by a macro-level example: consider the development of feathers by the dinosaur lineage ancestral to birds. Since it now seems that Archaeopteryx was either flightless or a very poor flyer, judging among other things from the architecture of its shoulder-girdle, and yet was fully feathered, it can be argued that feathers originally developed for something else. One persuasive view is that this was to serve as a thermoregulatory device for warm-blooded proto-birds living in high latitudes; this development was later opportunistically capitalized on or co- opted for flight (Bakker, I975; Ostrom, 1979). Exaptation then is the opportunistic co-optation of a feature whose origin is unrelated or only marginally related to its later use. In other words (loosely) a 'conceptual novelty' or 'invention'.

One consequence of this view is that organisms - and, I suggest, other historically evolved systems, like languages - may in their structure show a certain amount of bricolage; they are to some extent jury-rigged or cobbled together, and the remnants of old structures can be recobbled into new ones

8o

EXAPTATION IN LANGUAGE EVOLUTION

(cf. Jacob, I983, on 'molecular tinkering'). Konrad Lorenz (1978: 25) has a lovely image of an organism growing by evolutionary accretion: it starts off like a pioneer's shack, which serves as the nucleus of the grand new structure gradually erected around it. In the course of time the original shack becomes a kind of junk-room, and eventually almost every room in the growing house gets used for some non-original purpose. An organism is 'eine Menge Baumerkmale, die Uberbleibsel einer "Anpassung von gestern" sind' ['an aggregation of structural features that are leftovers from " yesterday's adaptation"']. A typical exaptation is the redeployment for a new purpose of one of yesterday's adaptations.

Gould and Vrba's important methodological point is that a totally selectionist evolutionary theory is constraining and heuristically unpro- ductive. By not insisting on the 'utility' of all parts of an organism, but allowing for 'nonaptations', features with no synchronic function, not doing anything, they permit organisms the freedom to evolve: 'the path of evolution - both the constraints and the opportunities - must be largely set by the size and nature of this pool of potential exaptations' (I982: I3).

I want to apply this kind of thinking to language change. Rather than viewing languages in the classical structuralist way as systems where (almost) tout se tient, let us consider the possibility of looking at them as Gould suggests we do at organisms: as 'bundles of historical accidents, not perfect and predictable machines' (I983: ioi). Language history may be largely a matter of 'mosaic evolution'; as Gould says in a biological context, 'an animal's parts are largely dissociable, thus permitting historical change to proceed'. I suggest that organic exaptation has a linguistic parallel, which may throw some light on the strategies languages use in their development. The idea may allow us to recognize a common strategic thread running through well known but previously not related or relatable phenomena. I think there is a reasonably common kind of linguistic change occurring in diverse guises, that can insightfully be seen as a sort of exaptation. I will use the term, with a certain benign looseness, to highlight a particular kind of historical scenario. (For some methodological comment see Section 4 below.)

My primary illustrations will be two examples from the history of Germanic; these have neither generally been recognized as odd, which they certainly are, nor as similar, which they can be argued to be. The first involves redeployment of the morphological exponents of an original aspectual opposition as markers of number concord; the second the downgrading of a syntactic contrast to mark the morphophonological properties of certain stem-classes.

A simple abstract case, which will be fleshed out in the next section, can serve as an illustration of the general principle. Say a language has a grammatical distinction of some sort, coded by means of morphology. Then say this distinction is jettisoned, PRIOR TO the loss of the morphological

ROGER LASS

material that codes it. This morphology is now, functionally speaking, junk; and there are three things that can in principle be done with it:

(i) it can be dumped entirely; (ii) it can be kept as marginal garbage or nonfunctional/nonexpressive

residue (suppletion, 'irregularity'); (iii) it can be kept, but instead of being relegated as in (ii), it can be used

for something else, perhaps just as systematic. (The new uses may be purely structural or intralinguistic, which is my main

concern here; or they may have a pragmatic/sociolinguistic dimension: see Section 5 below and note 13.) Option (iii) is linguistic exaptation. The point of course is that it is an option: languages may operate 'wastefully', dumping material that no longer does anything useful, or in a 'conservationist' mode, by recycling. This might in fact be a useful parameter for the typology of change.

As presently laid out, the concept of exaptation is not really precise enough; in particular I do not want to claim that ANY change in the use of linguistic material can be seen as exaptive, which would reduce the concept to triviality. First of all, prior coding of the category in question is not a precondition for exaptation: junk can be used for anything, since categories as well as uses for old material can be invented more or less ex vacuo. Second, there are many cases where old material is used for new categories that are not exaptive (this is the case with virtually all analogical transfers, for instance). Perhaps a clear example of nonexaptive re-use or extra use of existing, non-junked material will help to clarify the situation. The development of the modern Finnish 'local' cases shows what seems to be nonexaptive morphological invention. Proto-Uralic had a rather diffuse locative case with an */-n-/ formative, which survives in Finnish in reduced function as the essive, coding 'time-in-which' and the like (td-nd pdivd-nd '(on) this day') and certain locations (luo-na 'at the house of') and temporary states (poika-na 'as a boy'). More specific locations in Finnish are now coded by new cases, like the inessive (interior locative) or adessive (exterior or surface locative). These cases in Balto-Finnic have been constructed by means of the old locative preceded by what is called a 'co- affix': so *-na/-nd remains as the essive, but the inessive is -ssa/-ssd < *-s- na/-nd, and the adessive is -lla/-lld < *-Ina/-l-nd. The coaffixes however are not junk; the -1- formative in the adessive, to take one example, remains in the derivational suffix -la 'place where', as in ravinto 'food', ravinto-la 'restaurant'. There is no genuine 'novelty ', only extension of use, within the same semantic domain. (For details of the history of Uralic case systems, see Collinder I960: ?? 875-93.)

82

EXAPTATION IN LANGUAGE EVOLUTION

2. FROM SEMANTICS TO CONCORD: THE STRONG VERB PRETERITE

One of the great innovations characterizing Germanic is the destruction of the Indo-European aspect system. This was replaced by a tense-system, with no grammaticalized aspects. In rough outline, the IE present continues as the Germanic present, while the two 'past' categories, perfect and aorist, merge to form a new, conflated preterite:2

(I) Indo-European Germanic

PRESENT- PRESENT

PERFECT PRETERITE AORIST

Germanic also invented a new type of verb, the weak verb, in which the preterite is marked by a suffix containing a coronal (usually a stop); the other Germanic verb type, the strong verb, continues a type of tense-marking that has - as we will see - certain connections with the older IE aspect system.

In the weak verb, the aorist/perfect merger is complete, both semantically and morphologically, by the earliest records; all the original morphology that coded both categories has been lost. Consider for instance an early Northwest Germanic preterite like talgidai 'I carved' (N0vling clasp, c. 200

A.D.: Antonsen 1975: 30). This shows no sign of IE perfect or aorist morphology: it consists of a root with inherent o-grade vocalism (/taly-/ < */dolgh-/: cf. Skr daldyati 'he splits', L dolare 'hew'), plus the usual weak preterite machinery: root + thematic vowel + tense-suffix + personal ending, i.e. /taly-i-d-ai/. The only original material remaining is the root; the rest is innovative.

The strong verbs however were more conservative, and retained - if in a drastically altered capacity - much of the original contrastive morphology. To see what happened, we have to look back at one important pattern of morphological coding of aspect in Indo-European: ablaut. Many verbs showed a standard pattern in which different vowel-grades of the root were associated with particular aspects: thus, the e-grade marked present, the o- grade perfect, and the zero-grade aorist - whatever other marking might be present, such as reduplication, prefixed augments, etc. A representative

[2] Earlier scholars tried, often desperately, to derive the whole strong preterite from the IE perfect; for technical discussion and the standard arguments for the role of the aorist, as well as a good survey of the earlier literature, Prokosch (I938: 56 f.). Whether the IE aspect system taken as a base is 'original' or a secondary development is not germane; if the kind of system that shows up in Greek, say, is a late development, the loss of aspect in Germanic is loss of a secondary opposition, part of a cyclic pattern of loss and replacement (Lass, I987: ch. 6).

83

ROGER LASS

example might be the Greek present indicative active i sg lei'p- 'I leave', perfect ke-loip-a, aorist e-lip-on. The 'abstract' root is clearly /1_ip/; thus e- grade = /ei/, o-grade = /oi/, and zero-grade = /i/. The three root-forms are therefore /leip-/, /loip-/, /lip-/. The last is a 'zero' grade in the sense that the nuclear vowel has deleted, and the remaining /-i-/ represents -

morphophonemically - not a 'vowel' but part of the root. (For a perhaps more transparent example of 'genuine' zeroing cf. Latin pater 'father', nominative singular, with deletion of the second vowel in oblique cases, such as genitive singular patr-is.)

In addition to the zero-grade aorist there is another common type, which shows lengthened (usually e-) grade, and appears for instance in an important class of Latin perfects which stem from old aorists (cf. Buck, I933:

? 4I3). Examples are present e-grades ed-o 'eat', ven-i-i 'come', leg-o 'read', perfects ed-i, vin-T, leg-T. (These are distinct historically from secondary lengthenings in velar-final stems with old sigmatic aorists like rEx-T < */re:g- s-i:/, and stems in /-n/ like man-s-T, presents reg-o 'rule', mdn-e-o 'stay'.)

The important thing for our purposes is the existence in Indo-European of at least one major aspect-marking pattern in which present has e-grade of the root, perfect. has o-grade, and aorist has either lengthened e-grade or zero- grade:

(2) PRESENT PERFECT AORIST

{e:}

With this as background, consider the vocalic patterns of the Old Germanic strong verb classes I-III, here illustrated from Gothic and Old English. PRES = present system, exemplified by the infinitive, PRET1 = preterite I, 3 singular, PRET2 = preterite 2 singular and plural (for this usage and discussion see Lass & Anderson, I975: Ch. I):

(3) PRES PRET1 PRET2 I ' bite' Go beit-an bait bit-um

OE bit-an bat bit-on II 'bid' Go biud-an baup bud-um

OE beod-an bead bud-on III 'help' Go hilp-an halp hulp-um

OE help-an healp hulp-on The import of these forms may be unclear to the non-Germanist reader;

some etymological comment should make the historical relations apparent. PRES. Gothic ei = /i:/; both Go, OE /i:/ go back to IE */ei/ (Go steigan,

OE stfgan 'ascend' = Gr stei7ckd). In Class II, Go iu, OE eo reflect IE */eu/ (biudan, beodan = Gr peu'tomai 'inquire'). In class III, Go i represents a general raising of IE */e/ (cf. Go itan 'eat', L edo). Thus the PRES roots reconstruct as having nuclear */ei/, */eu/, */e/ in these three verbs.

PRET1. Except for the ea in OE healp, which represents a special

84

EXAPTATION IN LANGUAGE EVOLUTION

development before /lC/, all the PRET1 roots contain obvious a (probably a back vowel, but that is not relevant here). Germanic */a/ can reflect either IE */a/ (Go akkrs 'field' = L ager), or */o/ (Go ahtau 'eight' = L octo). Here the correspondence is to */o/, which allows reconstruction of the roots as */oi/, */ou/, */o/. OE a represents the normal development of */oi/, as in wat 'he knows' (cf. Gr oida, Go wait).

PRET2. If the basic roots in classes I and II are of the shape */Vi/, */Vu/, then the PRET2 -i- and -u- can be seen as residues, what one would expect from deletion of the nuclear vowels. In this sense bit- say is exactly parallel to Greek lip-: the vowel that appears on the surface is morphonemically part of the root and not part of the system of ablaut alternations per se - though by virtue of its remaining after deletion it serves as a secondary index of zero-grade. (There are parallels with -u- as well: Gr pheago 'I flee', aorist e-phug-on.) Class III is a slightly different matter, but the same in principle. Germanic /uR/ (where R = nasal or liquid) typically reflects an IE pre- sonorant zero-grade, where the sonorant is syllabified: thus Go wulfs, OE wuif 'wolf' < */wlkW-o-s/ (cf. Sanskrit vrk-d-h). Thus the -ul- in class III PRET2 is also a zero-grade relic, if of a slightly different kind.

At this point then PRES in classes 1-111 reconstructs with nuclear */e/, PRET1 with nuclear */o/, and PRET2 with zero; i.e. they reflect the old present/perfect/aorist alternation with some precision. The problem is that the PRET1 - PRET2 alternation does not correlate with tense (the reflex of IE aspect), but with number. Yet the PRET2 zero-grade is on the face of it unlikely to reflect anything but an aorist. Aside from anything else, it's more parsimonious to keep everything in the family. And since there's nothing in Germanic (or IE for that matter) historical morphophonology to support an alternation /a/ - /u/ (or /o/) 0 as a primary marker of number, the only reasonable source for the whole vocalic pattern is the old verbal ablaut. This is of course a handbook commonplace, since the strong verb classes are typically referred to as Ablautsreihen or 'ablaut series'; but the argument is rarely if ever made fully explicit, and it is even less often made clear that 'ablaut' is being used in a specifically Germanic, not a general Indo- European sense. At any rate, if what I am claiming so far is sound, behind the class 1-111 alternations lies an early Germanic archetype like this:

(4) PRES PRET1 PRET2 I -eiC- -aiC- -iC- II -euC- -auC- -uC- III -eRC- -aRC- -uRC-

And this in turn reflects a pre-Germanic archetype:

(5) PRES PERF AORIST I -eiC- -oiC- -iC- II -euC- -ouC- -uC- III -eRC- -oRC- -RC-

85

ROGER LASS

Class I then is directly equivalent to the old alternation pattern seen in Greek leip- - -loip- -lip-, and so on (the labels 'I-III' for IE do not of course reflect a 'system' of verb classes but simply their origin types). The first three classes then continue a classic ablaut series */e - o - 0/, with later local changes giving the attested root shapes.

Classes 1-111 have roots with a heavy syllable, i.e. rhymes in /-WC/, /-VCC/; the situation is different with classes IV-V, which continue a type with a light root syllable.3 These also show what we can now recognize as an o-grade perfect in PRET1, but have a lengthened e-grade in PRET2: given the reasonably unambiguous pairing of perfect and aorist in 1-111, and the clear o-grade perfect in I-V, this is a straightforward interpretation.4 Examples:

(6) PRES PRET1 PRET2 IV 'bear' Go bair-an bar ber-um

OE ber-an bir bkr-on V 'eat' Go it-an at at-um

OE et-an at tt-on

Gothic <ai> = [F], a predictable reflex of */e/ before /r/; Go /e:/, Oe /m:/ continue IE */e:/, as in L ed-T, itself as we have seen an old aorist. We can now add to the archetypes (4-5) the ones for classes IV-V:

(7) Germanic Indo-European

PRES PRET1 PRET2 PRES PERF AORIST IV -eR- -aR- -e:R- -eR- -oR- -e:R- V -eC- -aC- -e:C- -eC- -oC- -e:C-

The development from Indo-European to Germanic can be diagrammed as follows (class III as exemplar: only the phonological details will vary from class to class):

[1] I omit strong verb classes IV and VII, as their historical root structure (probably containing a laryngeal) does not allow the old patterns to surface clearly. Both show the same (long) vocalism in PRET1/PRET2, which makes them historically uninformative. A typical case is class VI 'bake', OE PRES bacan, PRET1 boc, PRET2 bocon, where both length and quality distinctions have been lost. This could in fact be referred to an old pattern like */bHeg- - bHog-/, etc.; but the arguments are too complex to go into here (see Lass & Anderson, I975: 49 ff., where we try to resurrect the laryngeal as a synchronic abstract segment).

[41 It has been suggested that the zero-grade ancestors of class III PRET2 in fact have a perfect, not an aorist origin: there is a (less common) zero-grade perfect plural type, as in Sanskrit kar- 'make', perfect active I singular ca-kar-a, plural ca-kr-md. This is taken as the source by Krahe (I963: ? 6i). Certainly this is possible; but since an aorist is needed for the lengthened grade of classes IV-V anyhow, and the whole system derives from an aorist/perfect syncretism, my scenario is simpler. However, the fact that the zero-grade perfect is associated in some cases with plural (if this was indeed the case in the IE dialects ancestral to Germanic) may of course have a part to play.

86

EXAPTATION IN LANGUAGE EVOLUTION

(8)

Indo-European: PRES - eRC - PERF - oRC - - RC - AORISTI

I PRET Germanic: PRES - eRC - PRET1 - aRC - -uRC- PRET2

The scenario encapsulated here would appear to be: (i) loss of the (semantic) opposition perfect/aorist; (ii) retention of the diluted semantic content 'past' shared by both (in

other words, loss of aspect but retention of distal time-deixis); (iii) retention of the morpho(phono)logical exponents of the old categories

perfect and aorist, but divested of their oppositional meaning; (iv) redeployment of the now semantically evacuated exponents as

markers of a secondary (concordial) category; in effect re-use of the now 'meaningless' old material to bolster an already existing concordial system, but in quite a new way.

After the merger (i), the remnants of the perfect/aorist ablaut system, divorced from the opposition they instantiated, are historical junk; they are pressed into the service of a new linguistic function (which was in fact always coded in any case, if mainly by suffixes), rather than being dumped altogether. (The dumping of the system comes later, when the redundant vowel-grade+suffix coding of plurality is lost in most Germanic dialects, with retention only of suffix as in German and Dutch, or complete loss of number marking as in continental Scandinavian, English and Afrikaans. Only Icelandic now maintains the distinction, as in class I bi'ta 'bite', PRET1 beit, PRET2 bitum, class II bjf6a 'offer', bau6, bubum, etc.)

What is of prime interest here is conceptual novelty: ablaut in the Indo- European sense was never used for this kind of thing before. Indeed, given its original conditioning factors (position of accent, major morphosyntactic category, etc.), it never could have been. With the breakup of both the original aspect system and the function of ablaut itself, the old forms were available for cooptation.5

[51 The redeployment of morphology after a category breakdown does not of course have to be systematically exaptive. When the IE nominal ablaut deteriorated, there was in certain declensions a near-random redistribution of formerly conditioned vowel grades, often on a lexeme-by-lexeme basis. So the word 'tooth' in Germanic shows both o-grades (OE top < */tanO/ < */dont-/, OHG zand, OIc tQnnr) and zero-grades (Gothic tunAus < */tun0/ < */dnt-/. For discussion see Lass (I986). Similar phenomena can be seen in the survival of IE locatives, instrumentals and datives in the syncretic Germanic dative (Lass, forthcoming).

87

ROGER LASS

3. FROM SYNTAX TO (MOSTLY) MORPHOLOGY: AFRIKAANS

ADJECTIVAL -e

My second case is not so much one of massive exaptation of an old system as a new and conceptually innovative form of inflection, with no concordial (and virtually no semantic) function. But it is still exaptive, in that the surface exponents of an old contrast - if in rather degraded form - are retained, and pressed into a quite new service.

All Germanic languages except English show some remnants of the Old Germanic adjectival inflection (English did until about 1400 - see Mustanoja, I960: 275 ff.). Despite complex local developments, such as a dual declension ('strong' vs. 'weak'),6 the principles are more or less the standard ones of Indo-European adjectival inflection: the adjective is an 'empty' category with no inflections of its own, all its marking is concordial. The concord triggers are syntactic: features of the whole NP such as definiteness or quantification, or features of the head noun like (surface) case, number and gender.

The historical background of the Netherlandic adjective system is obscure; in particular we have only fragmentary attestation of the declension in the one corpus that could be called 'Old Dutch' -the Old Low Franconian (OLF) psalm glosses and laws.7 But the general inflectional principles of the older-style dialects can be illustrated clearly enough with OLF examples (texts from Markey, 1976: Psalm I8: 4, 6):

(g) (a) an all-ero erth-on to all-FEM. DAT. SG. earth-WEAK FEM. OBL.

'to all the earth' (b) fan ho-on himil-i

from high-MASC. DAT. SG. heaven-MASC. DAT. SG.

'from (the) high heaven'

By Middle Dutch, the strong/weak distinction had been lost (cf. van Loey,

[6] The strong declension has a rich paradigm, the adjective carrying some marking for case, number and gender (cf. modern German gut, -er, -es, -em, -en etc.). This tended to appear in collocations with no predeterminer, or where the determiner itself was inflectionally equivocal or unexpressive. The endings derive mainly from nominal or pronominal case/number suffixes (e.g. German -es for m/n genitive sg). The weak declension was much less highly differentiated, and borrowed its morphology from the weak n-stem nouns (cf. German des gut-en Mann-es, dem gut-en Mann). This was used most often where the concordial/categorial information was carried by the determiner, and the adjective inflection marked mainly nominative vs. oblique. The intricacies of this system (which varied enormously from dialect to dialect anyhow) will not concern us here.

[7] There are a few other fragments as well: one famous West Flemish sentence in an eleventh- century English MS (Sch6nfeld, 1932), and a handful of others (Sanders, 1972). Old Low Franconian (ninth-tenth centuries) is the only substantial evidence for a 'Proto- Netherlandic' ancestor of the cluster of text traditions arising in the late twelfth century and conventionally called 'Middle Dutch' (cf. van Loey, I970a; Raidt, I980: Chs. 3-4). This complex cluster, with its fuzzy antecedents, is the ancestor of the equally complex and heterogeneous 'seventeenth-century Dutch' that formed the input to Afrikaans.

88

EXAPTATION IN LANGUAGE EVOLUTION

197ob: 269f.). The adjective often took its oblique form from the inflected article, and both endingless and inflected forms could be used in the same context. We can however extract a generalized paradigm for attributive adjectives of this type:

(I0) m n f pl NOM -e -e -e -e GEN -(e)s -(e)s -er -er DAT -en -en -er -en ACC -en -e -e -e

This is fairly impoverished (compare for instance what was available in Old English or Old High German, or Modern Icelandic still). But it was also subject to a certain amount of variation in use; for instance, old weak genitives (in fact homophonous with datives and masculine accusative singulars in the strong declension) and strong genitives could be used in the same context: des goed-en/goet-s mans 'the good man's'. But despite this messiness and variation, and even though the system (or lack of it) is quite innovative with respect to earlier Germanic models, it is still - and this is the vital point - morphosyntactically based. The import of this will be clear shortly. Even in modern Dutch, where the control of adjective inflection is wildly different in detail from anything that was possible in Old or Middle Germanic dialects, the triggers for adjective inflection are still roughly of the same type. That is, the controllers are by and large features of the head.

So much for background. My interest here is in the specific developments in Afrikaans, which are quite late, beginning no earlier than the mid- seventeenth century. I will therefore skip over the decay and transformation of the pre-seventeenth-century inflection, and focus on the pattern that Afrikaans eventually deviates from, and the nature of that deviance. For seventeenth-century Dutch, despite an enormous amount of variation,8 we can extract some fairly clear principles, not too unlike those still in operation. The old three-gender system had broken down, and collapsed to a two-way opposition 'common' vs. neuter, signalled primarily by the definite article: common de vs. neuter het, as still in standard Dutch. (On the history of Dutch gender see Dekeyser, I980.) Aside from survivals of the old genitive and dative inflections in certain instances, the adjective was essentially either endingless or in -e.9 Since the Afrikaans system is built entirely on the opposition Adj-0 vs. Adj-e, I will restrict my remarks to this.

[8] It is hard to specify just what the Afrikaans input was like, since there were many regional and social varieties of the not-yet-standardized seventeenth-century Netherlandic complex involved. There is little point here in trying to reconstruct a systematized Early Modern Dutch morphology; I give below only enough details to make the main points. Modern standard Dutch shows an essentially cognate and conservative system. For the Afrikaans input see Raidt (I983); another two-way zero vs. -e system can be seen in Frisian.

[9] The modern system is very much the same: zero vs. -e except for relics of old datives and genitives in lexicalized expressions and archaizing styles (see Rijpma & Scheuringa, I969:

? I17).

89

ROGER LASS

The primary controls on the presence vs. absence of -e were (a) the gender of the head noun (neuter favoured lack of -e), and (b) the nature of the determiner (if any): indefinite and quantified nouns favoured -e if common and zero if neuter. Definites were (and still are) variable: preferentially neuters are endingless, and commons take -e, as do (tendentially) plurals of both genders.

Some characteristic seventeenth-century examples (taken from texts quoted in Raidt, I980: Ch. 6) will illustrate the commonest types:

(i i) (a) Common Sg in een lang-e ry 'in a long row' hoog-e waaternoodt 'great water-famine'

(b) Neuter Sg ons dagheliks broed geeft ons 'give us our daily bread' een zwart mantelken 'a little black cloak'

(c) Plural de groot-e huizen 'the big houses' (neuter) onduitsch-e termen 'un-Dutch terms' (common)

The system was obviously more complex and subtle than this, but the neuter/common and indefinite/definite as well as singular/plural contrasts make the point. The conditioning of -e vs. -0 is purely morphosyntactic (or lexico-semantic in the case of gender); but even when an 'internal' property of a noun, like gender, is involved, it is the syntagmatic or 'external' environment that determines the shape of attributive adjectives. Predicate adjectives in any case are endingless.

In early Afrikaans (Kaapse Nederlands, Proto-Afrikaans or whatever one wishes to call the language from I652 to about I750), the Dutch system collapsed completely, and was replaced by something totally novel. The first phase of this process was loss of grammatical gender. The erosion of the de/het distinction was already under way in the seventeenth century, both de and het giving way to the generalized die by about 1740. The reorganization of the adjective to be described was complete by around I775 (Raidt, 1983: 1490.

Now it would seem likely that once the basic trigger of gender was lost, the distribution of -e would for a time be close to random; each adjective 'had' a form in zero and one in -e, and the absence of gender-specification should allow reasonably for either one surfacing in a given context. That is, one might expect indefinite neuters in -e like een kleyn-e stuk 'a little piece', witte- water 'white water', and zero-ending commons like een ander plaats ' another place', een yzer harpoen 'an iron harpoon', alongside the 'correct' types een kleyn N (neuter), een ander-e N (common). And indeed, in the 'transition' period between the Dutch and Afrikaans systems, this is precisely what we do find ('deviant' forms cited from Scholtz, I98I: 129, which gives an excellent historical overview). This is the crucial 'junk' stage.'

90

EXAPTATION IN LANGUAGE EVOLUTION

This sort of distribution should, if we go by the example of similar cases like that of late Middle English, lead to a stabilization of one form or the other, i.e. loss of inflection. Once the primary conditioning factor is gone, we would expect the exponents of the contrast to go as well - especially since there is good documentary evidence for a period of extremely messy and 'senseless' (apparently non-deterministic) variation. But this is not what happened. Afrikaans not only did not lose the -e/-0 contrast, it restabilized it and redeployed it in a new and complex (and more rigid) system, at a different grammatical level. The now 'baseless' opposition was co-opted in the service of a conceptually novel inflection-type, with no real Germanic precedents. In particular, the new system is non-syntactic: definiteness, number, quantification, presence or absence of predeterminers, case (and of course gender) play no role. I will look at the new system in detail below; for now it suffices to give a summary of the kind of development that could be claimed to lead up to the present state of affairs:

(12) Junk-deployment in Afrikaans: idealized scenario

I. Gender marked morphologically:

een kleyn-e Lharpoen een kleyn- [stuk common neuter

II. Loss of gender

III. -e is now junk: een kleyn - kleyn-e harpoen

IV. Adapt or die: -e saved by redeployment (exaptation)

V. Result: any adjective is inflected or not as a categorial property, regard- less of the head noun it modifies, or any NP features.

It now remains to flesh out the new system. Following the basic analysis in Raidt (I983: I84 ff.), but not her specific taxonomy, we can describe the new system this way. First and foremost, the domain for inflection is THE

PARTICULAR ADJECTIVE ITSELF, NOT ITS SYNTACTIC ENVIRONMENT. A given adjective is generally either inflected or uninflected in all attributive contexts. There are two major classes, then, categorically inflecting and categorically non-inflecting.

9I

ROGER LASS

3. I . Categorically inflecting adjectives

These may be divided into two main groups, (i) morphologically complex adjectives, and (ii) monomorphemic but morphophonemically complex adjectives.

(i) Morphologically complex. With one exception to be noted below, any polymorphemic adjective takes attributive -e: so in this group are prefixed items like ge-heim 'secret', suffixed ones like stad-ig 'slow', compounds like open-baar 'public', prefixed compounds or quasi-compounds like be-lang-rik 'important', etc. The syntax of the NP is irrelevant: so 'n geheim-e resep 'a secret recipe', geheim-e resepte 'secret recipes', baie geheim-e resepte 'many/very secret recipes', die/hierdie/geheim-e resep(te) 'the/this/these secret recipe(s)', etc. The only relevant syntactic feature, here as elsewhere, is that inflection occurs only in attributives: predicate adjectives are endingless, as in die resep(te) is geheim 'the recipe(s) is/are secret'.

The main exception is also morphologically conditioned: comparatives are endingless, regardless of whether the base inflects or not: 'n geheim-er resep 'a more secret recipe', 'n groot boek 'a big book '/ 'n groot-er boek 'a bigger book'. Comparatives in fact fall in with obscured complex adjectives like ander 'other', lekker 'delicious', which are non-inflecting (obscured in the sense that -er is not perceived as a suffix, even if historically it might have been).

(ii) Morphophonemically complex. The next cut is between those adjectives with variant stem-allomorphy (complex) and those with only one stem- allomorph (simple). Adjectives with only one stem-allomorph (if they do not fall under (i)) are non-inflecting; those with alternants take -e. As will become obvious, this is rather a chicken-and-egg set-up; it is only the preservation of the inflected forms that allows the alternations to persist, since they are all conditioned by final vs. non-final position. The main groups are:

(a) Cluster-simplification alternators. Dutch already had a tendency to simplify some foot-final obstruent clusters, as can be seen in vos, os compared to their English cognatesfox, ox. Afrikaans has carried this further, deleting (inter alia) the stop in most coda-clusters containing a stop and a fricative (in either order): thus pos 'post' vs. Dutch post, plaas 'place, farm' vs. Dutch plaats, and so on. (There are exceptions, e.g. rots 'rock ',fiets 'bicycle', many of which are late loans.) In Afrikaans the deletion occurs only if the cluster is absolutely final in the foot, that is if both obstruents involved belong to the coda of the same (strong) syllable. Thus vas 'fast', sag 'soft', reg 'right', inflected vast-e, sagt-e, regt-e, i.e. /fas - fast-a/, /rex - rext-a/, etc. The role of syllable position can be shown as follows (these representations are indifferently either historical or synchronically 'underlying', as you wish):

92

EXAPTATION IN LANGUAGE EVOLUTION

(I 3) vas vaste reg regte

BASE [fast] [fa[st]a] [rext] [rex[t]a]

DELETION 0 In vaste the /st/ cluster is ambisyllabic, since /st/ is a permissible syllable- initial; in regt-e the /xt/ is partially split by a syllable boundary since /xt/ is not a legal initial. But in both cases the cluster as a whole is not uniquely in the coda of the first syllable in the disyllabic form, which is the environment for deletion. (On the syllabification model invoked here see Lass, I984: ? IO.3.5f.; Anderson & Ewen, I987: ? 2.3.)

(b) Medial syncope alternators. In Dutch, and to a larger extent Afrikaans, single consonants (preferentially /d/ and /x/ < /y/) tend to delete medially in the foot, that is when they are NOT exhaustively in the coda of the strong syllable, but form an ambisyllabic 'interlude'. Examples are weer 'again' (cf. German wieder), voil 'bird' (cf. Dutch vogel). This syncope produces alternations in adjectives that retain final -e: droog /droax/ 'dry', inflected droi /droa/, oud'old' /ceut/, inflected oue /cu;)/. Syncopated /d/ usually leaves a palatalized residue /j/ behind, as in goed /xut/ 'good', inflected goeie /xuja/, so also dood /doat/ 'dead', inflected dooie /doaja/. In cases where the /j/ does not appear, the syncopated form may be relexicalized, as in ouer 'older' (cf. Dutch ouder), leading to ouers 'parents', ouerdom '(old) age'.

Some items in this group tend to have their syncopated forms lexicalized as full new attributives without -e; two common ones are vroi 'early' - vroeg, dooi 'dead' -, dood. The status of these forms is uncertain, but they appear to be colloquial (probably nonstandard).

(c) Auslautverhartung alternators. As a typical continental West Germanic language, Dutch shares with German and has bequeathed to Afrikaans a medieval rule of final obstruent devoicing - both the categorical effect of the rule on surface phonotactics, and a set of morphophonemic alternations involving it. Typical instances are blind /blint/ 'blind', inflected blinde /blfnda/, hard /hart/ 'hard', inflected harde /harda/, doof /doaf/ 'deaf', inflected dowe /doava/ (cf. non-alternating hart 'heart' /hart/, plural harte /harta/). The representations here are surface-phonemic in the simplest sense, not 'underlying'; Afrikaans spelling apparently vacillates, as we can see from morphophonemic hard/harde vs. phonemic doof/dowe.

There are a few other items in the alternating group that do not belong specifically to these major classes: a typical example is nuut /nyt/ 'new', inflected nuwe /nyva/.

93

ROGER LASS

Taking this alternating group as a whole, it is interesting to note, as a theoretical sidelight, that their history presents a clear case of change acting to MAXIMIZE allomorphy, in the most straightforward possible way. It would have been so 'easy' to drop the form in -e (as was done with so many other adjectives). It is curious that in a language that has done so much radical hatchet-work on its paradigmatic structure (losing all its verb inflections for person and number, for instance, and obliterating the nominative/ accusative/genitive distinction in the first and third person plural pronouns), the simplification (or 'simplification') of the adjective declension should have led to the preservation of a host of minority paradigm-types.

3.2. Categorically non-inflecting adjectives

The bulk of these are monosyllabic, ending in obstruents like diep /dip/ 'deep', los /los/ 'loose', vowels like vry /frEi/ 'free', blou /blceu/ 'blue', sonorants alone like geel /xeal/ 'yellow', or sonorants in clusters like dronk /drosjk/ 'drunk'. A few of these non-inflecting adjectives are polysyllabic, either in -er like ander, lekker (see above), or comparatives. Raidt gives a detailed structural survey of this class, but it is really only necessary to specify two characterizing features: (a) morphologically simplex (non-comparative), and (b) having only one stem form. This class then is by and large a residual default or 'elsewhere' category.

A few adjectives, mostly in -el, do not fit comfortably into either class, but display another kind of exaptive strategy. In these, the domain is semantic: two related meanings may be differentially coded by presence or absence of -e. A good example is enkel 'single'. In collocation with man 'man' it can have two senses: 'n enkel man 'a solitary man' vs. 'n enkel-e man 'a single (unmarried) man' (thanks to Melinda Sinclair for this example). This is a marginal semantic exaptation, which is not widespread; other -el adjectives seem to vary without semantic consequences, though the conditioning - if any - is obscure. It does seem however that -el, unlike -er, rather favours -e. It may be that -el is optionally interpretable as a kind of pseudo-suffix, which would allow enkel to be interpreted either as enkel or enk-el (on the model of the nominal -(s)el in words like stelsel 'system', mengsel 'mixture' - cf. stel 'set, suite', meng 'alloy, blend, mix').10

So the rules controlling -e (categorical or variable) crucially involve the following:

(a) morphological structure (simple vs. complex); (b) paradigm-structure (alternating vs. non-alternating); (c) lexico-phonological idiosyncrasies (-er disallowing -e regardless of

morphology, doublets like those for enkel).

[io] It is worth noting that the domain of exaptation does not have to be morphosyntactic (though here it is at its most systematic). Semantic splitting of doublets (whatever their origin), as in person/parson, kirk/church (in Scotland), skirt/shirt, and the like is also clearly exaptive: if two forms code one meaning, one form is (potential) junk.

94

EXAPTATION IN LANGUAGE EVOLUTION

This throws into relief the conceptual novelty: the system here is as far as I know unique in Germanic. All information controlling addition or non- addition of -e is internal to the adjective itself as a word-form or to its morphophonemic paradigm.11 In the earlier (syntactically based) inflectional system, the information controlling suffixation was essentially syntagmatic: nature of determiner, features of head noun, and so on. The exaptation here is the 'internalization' or 'introjection' into the adjective itself of the inflectional triggers, using the inherited, formerly syntactically controlled -e purely 'locally', as a marker of adjective class. Globally, one could say that the inflectional locus has shifted from syntax to lexicon. (That is, on the assumption that morphophonemic alternations are lexically coded, not produced by 'abstract' derivations from underlying representations of the usual kind. On an abstract analysis class I adjectives would still exert lexical control, but this would be in terms of phonologically coded 'underlying' properties, not paradigmatic ones.)

The exaptive scenario begins with the loss of gender; when common and neuter nouns are no longer distinguished, the -e is free to be used for something else. In Middle English there was a somewhat similar but less innovative solution, which never bypassed the syntax; adjectival -e was first bleached of its old case/gender and definiteness/indefiniteness content, and finally exapted as primarily a marker of plurality. In the fifteenth century, partly as a consequence of a generalized loss of final /a/ (which did not happen in Dutch or Afrikaans), the -e was lost, and the adjective became invariable. English is thus 'wasteful' with respect to this piece of inherited structure, Afrikaans is 'conservationist' (cf. Section i).12

4. INTERLUDE: HEURISTIC METAPHOR VS. CRUDE TRANSFER

Readers with an interest in matters epistemological or methodological could be a little worried at this stage. Was I merely paying lip-service to my own caveats in Section I, and have I, in fact, done what I criticized the vulgar-Freudian literary critics and the like for doing? That is, have I taken a theoretical construct whole-hog out of its matrix (where it has a genuine theoretical interpretation) and reified and misused it? This is an important issue, because evolutionary and other biological concepts have on occasion

[iI] The structure of a paradigm may serve as a trigger for or a brake on change. For instance, one of the major constraints on final /-a/ deletion in Yiddish nouns was the pluralization class they belonged to. Most nouns in /-a/ lost it in late medieval times (e.g. harts 'heart', cf. OE heorte, German Herze); the main exceptions were those nouns belonging to the s- plural class, like kacke 'duck', pl kackes (< Polish kaczka). Other instances of retained /-a/ formed (parts of) morphologically non-root material, as in the diminutive suffix -ele (ges-ele 'little street'). For details, see Lass (ig80b).

[I2] The variation in some -el adjectives with respect to -e suggests that Afrikaans might in the end go the way of English, and level all adjectives to zero-inflection. Aside from noun plurals, adjectival -e is the only productive inflection (aside from the ge- prefix on most past participles, which always co-occurs with an auxiliary).

95

ROGER LASS

(at least since the days of Pott and Bopp: Wells, I987; Morpurgo Davies, I987) been transferred to linguistics in a rather gross fashion; and the worst examples, if taken as typical, might tend to discredit the enterprise in general.

Let me clarify what I have been trying to do by indicating what I am not doing. In no sense do I advocate a (serious) 'biological model' (Stevick, I963; Dimmendaal, I987) for language change. All I am doing is taking up what Mary Hesse (I966) calls the 'positive analogy' between biological and linguistic evolution in the exaptation metaphor, and leaving aside the (probably larger) 'negative analogy'. In other words, while claiming that the notion of exaptation seems useful in establishing a name and descriptive framework for a class of historical events, I remain fully aware (even insistent) that languages are not biological systems in any deep sense - even if they are used and transmitted (and were ultimately invented) by biological systems. The cultural/symbolic is ontologically distinct from the biological, not least by virtue of being non-physical. There is as far as I am aware no storage or coding mechanism for linguistic transmission equivalent to DNA. If I am saying anything of general import (and I think I may be), it is about properties common to historically evolving systems regardless of their substrates; rather than extending a notion from biology to linguistics, I am suggesting that the two domains (and others as well, probably, like the evolution of art styles or political institutions or sartorial fashion) have certain behaviours in common by virtue of evolving.

If, say, one were to conceive a language at a given time ti as a population of variants whose next state ti+1 is the result of 'selection' of certain individuals rather than others, one is not making a category mistake, and necessarily likening a language to a biological population/species in any way - insofar as mechanisms, mode of being and the like are concerned. In other words, given ANY population of individuals that show some variation (aspects of styles, constructions, genetic constitutions), and some (unspeci- fied) conditions that prevent all of them from surviving and predispose to the survival of certain individuals or types, the 'Darwinian' mode of talking becomes an appropriate one (cf. Dawkins, I982, on 'universal Darwinism'). And once this happens, the metaphor-space in which the original was conceived and now operates becomes an appropriate one to explore for further heuristics, taxonomic clues, and so on.

The point is worth making because there are writers who have not avoided the category mistake that I have tried -maybe in what looks like a pussyfooting way - to steer clear of. Perhaps the most striking recent example is the theory of 'linguistic paedomorphosis' proposed in a number of publications by Bernard Bichakjian (I984, I987, I989). Bichakjian starts from the now familiar idea that humans are somewhat neotenous or paedomorphic animals: they carry into adulthood morphological or other characteristics that are typical of the pre-adult stages of their ancestors, or

96

EXAPTATION IN LANGUAGE EVOLUTION

become sexually mature while still in what (historically) can be conceived as a partly 'larval' or 'foetal' state. Typical features in which human adults resemble foetal apes are their relative hairlessness, domed forehead, downward-facing foramen magnum, ventrally directed vagina, unrotated big toe, very high cranium-to-body-size ratio, and delayed skeletal calcification (cf. Gould, I977: Ch. 9; I983: Ch. 7 for discussion and references).

Bichakjian's transfer consists of the claim that linguistic evolution has an inherent bias for proceeding so that the outputs of change tend to move 'in the direction of earlier-acquired features and strategies' (I987: 99). For instance, he argues that the Indo-European languages have been steadily evolving toward a preference for right-branching rather than left-branching syntax, and that right-branching structures are typically acquired earlier. (I do not know if this is true or not, but his evidence is sketchy enough to raise doubts.) This claim is bolstered with an elegant blocking-clause to the effect that only 'inherent' rather than 'contingent' changes proceed this way, so that no counterexample can in fact be damaging (I987: 98 f.). He suggests (ibid.) that 'this ongoing evolutionary process, which closely resembles man's biological evolution', has in fact the same source as the physical manifestations: it 'should probably be traced back to the action of regulatory genes on the cellular correlates of speech'.

The problem is of course that we know nothing about 'the cellular correlates of speech'; indeed, it is not at all clear unless one happens to be a pretty devout Chomskyan that there are, or are likely to be, any such things. Further, since in its respectable form the theory of paedomorphosis is a theory of adult physical resemblances to non-adult physical forms of evolutionary ancestors (of course controlled by regulatory genes, since the whole thing is a matter of relative timing), the transfer to the cultural products of adult humans is at best bizarre, at worst irresponsible. No biologist has ever claimed that we THINK like foetal apes. The transfer here is crude and literal, based on terms and concepts that have proper and conventional interpretations in an ontologically specific (and highly different) domain; the transfer has been effected in the terminology (without the substance, but embodying a claim about substance) of that other domain.

It should be clear that I have avoided this kind of thinking, even to the extent of using terms like 'strategy', 'cooptation', etc. with no ontological specificity. Bichakjian's claims are scientistic, in that they purport to be hypotheses about (ultimately) physical evolution, but merely ape (!) the structures that biologists have erected on totally different grounds. Even many hard-nosed physicalist (neo-)Darwinian biologists would cheerfully admit that cultural evolution is extrasomatic. What I have attempted here is not 'hypothesis-formation' in the strict sense at all, but something much more modest: a kind of exegetical re-examination of certain types of linguistic change, with initially no higher aim than naming as a taxonomic

97 4 LIN 26

ROGER LASS

ordering device. While the claim is weaker, it is also safer, and may therefore be more useful. At least it escapes silliness, even if it is not a proper falsifiable hypothesis.

5. EPILOGUE: THE JOYS OF IDLENESS

I return now to some less philosophical concerns. The patterns of change discussed here display an important property of evolving systems (not only linguistic ones): useless or idle structure has the fullest freedom to change, because alteration in it has a minimal effect on the useful stuff. ('Junk' DNA may be a prime example.) Major innovations often begin not in the front line, but where their substrates are doing little if any work. (They also often do not, but this is simply a fact about non-deterministic open systems.) Historical junk, in any case, may be one of the significant back doors through which structural change gets into systems, by the re-employment for new purposes of idle material.

Crucially, however, mere 'uselessness' is not itself either a determinate precursor of exaptive change, or - conversely - a precursor of loss. Historical relics can persist, even through long periods of apparently senseless variation, and it is impossible to predict, solely on the basis of such idleness or inutility, that ANYTHING at all will happen. (That such predictions can be made if we get the right theory is an endemic delusion of historical linguists: see the arguments in Lass, ig80a: Chs. 2-3.)

Two examples, of somewhat different kinds, may be useful as illustrations of the range of acceptable behaviours. First, the fate of i-umlaut in various West Germanic dialects. This change (of ancient date) was provoked by suffixal */i, j/ which fronted the vowels of root syllables; when these suffixes were lost or neutralized, the vowel-alternations they produced were phonologized: so OHG gast 'guest', pl gast-i, MHG gast/gest-e, Modern German Gast/Gdste, etc. With phonologization came the possibility of morphologization: umlaut could be interpreted as an (opaque) mode of plural formation, with no phonological conditioning. At this stage, it could be treated essentially in one of two ways: either (since it was very much a minority type), it could be dropped in favour of one of the more 'regular' or more statistically common pluralization strategies; or it could be analogically extended to nouns which historically used a different strategy.

The first tack was taken by Netherlandic; in modern Afrikaans, for instance, virtually all the classical umlaut plurals have been reassigned to other declensions (pretty much as in Dutch): so German Gast/Gdste =

Afrikaans gas/gaste, Fuss/EFisse = voet/voete, Buch/Biicher = boek/boeke, etc. The only umlaut noun that appears to be in common use is stad 'city', pl stede. (Other plurals involving vowel-alternations have different sources, e.g. skip 'ship', pl skepe, which is a reflex of open-syllable lengthening and lowering, and irrelevant here.)

98

EXAPTATION IN LANGUAGE EVOLUTION

German on the other hand indulged to some extent in the second option: e.g. Baum 'tree', pl Bdume < OHG boum/boum-e. Yiddish did this even more extensively; not only native words, but loans from all periods could be reinterpreted as umlauting. So toxes 'arse', in addition to its (rarer) Hebrew plural txosim has the remodelled texes-er; and English sweat-shop was borrowed as the half-calque svits-s'op, with the second element subject to umlaut: p1 svits-sep-er.

English steered a kind of middle course: a large number of the old umlauting nouns were reassigned (book, nut, cow for instance now take -s plurals, instead of expected *beech, *nyte, *kye < OE bec, hnyte, cy); a small number remain (man, tooth, foot, mouse, louse); but no new ones have been added. (Though umlaut has in a sense been exapted as a minor joke-strategy, as in meese for pl moose, etc.)

Umlaut plurals are a slightly peripheral example, since they are not functionless or totally idle; still, they illustrate the general point about strategic options, which is an important one. A more central example of the persistence of genuinely idle and unexapted structure - but with some interesting possibilities - is the -s ending of the English present indicative 3 singular. In the course of its evolution, English has lost all the non-tense morphology of the verb except this suffix; it now has none of the 'informative' concordial function that verb suffixes had in earlier times (actually before it was, except in the North, -s, but was still the ancestor of -th). English once had a much freer word-order than it does now, and subject pronouns were not obligatorily expressed. Now however this relic inflection not only has no 'communicative' function in the sense of serving potentially to underwrite the parsing of ambiguous structures; it is a systemic excrescence. (Other languages with similar excrescences or asymmetries may clean them up; Afrikaans did by dumping all its verb morphology, and Swedish and Danish did by extending one ending - originally that of the present second person singular - to the whole paradigm.) But the exaptive impulse is strong; we might note that the -s ending can be exapted for sociolinguistic purposes, as an indexical marker of vernacularity in some dialects, where it is extended to all person number forms (see Cheshire, I982: Ch. 4 on this use in nonstandard English in Reading).

We should in fact not be surprised at the retention of historical junk over long periods. Despite neo-Praguian claims (e.g. Martinet, 1955: 49 f.: cf. Lass, ig8oa: 9I f.) that there is a kind of 'expense of energy' in the maintenance of oppositions that predisposes to loss of items with low functional load, there is really no evidence whatever that linguistic systems have 'thermodynamic' properties of this kind. It does not take any 'energy' (even the image is inept) to maintain historical residues, no matter how useless they may be at a given moment. The zero option is easy: do nothing, and the objects in question do no harm by just lying there.

The often bizarre and seemingly motiveless complexity of linguistic

99 4-2

ROGER LASS

systems is, like many of their other properties, simply a matter of historical inertia. Nonaptations persist because there is no particular problem in keeping them, and there may even be 'work' to do in getting rid of them. If these nonaptations or adaptations fallen into desuetude can be later exapted for something else, well and good; but there is no particular reason ever either to do this or not to.'3 Like so much else in so many domains, this is a matter of the 'creative' freedom available to historically evolved systems, precisely because of their enormous complexity and innate conservatism. As Lorenz remarks in his discussion of the junkroom-in-the-mansion alluded to in Section i, one of the reasons that ancient material persists is that it is very hard to dismantle a house while you are still living in it:

Die also solche erkennbaren historischen Reste bleiben schon deshalb erhalten, weil der Bau nie ganz abgerissen und neu geplant werden konnte: das war Gerade deshalb unmoglich, weil er dauernd bewohnt und intensiv beniitzt wurde.

['The recognizable historical remains are retained, because the structure can't be entirely torn down and planned anew; this would be quite impossible so long as it was being continually inhabited and intensively used'.] With luck, however, you can redecorate.

Author's address: Department of Linguistics, University of Cape Town, Rondesbosch 7700/RSA.

[I3] One of JL's referees objected strongly to this: 'it is really outrageous to say ... that the seemingly motiveless complexity of languages is a product of their historical inertia ... Since Weinreich et al. surely it has been pretty well accepted that alternation/variation of a systematically pointless type is an important mechanism of linguistic change'. I agree that it is; but to attribute it to a 'function' of being there to enable change to take place is to invoke backwards causation in the same way that Gould and Vrba carefully avoid doing in their definition of exaptation (see Section I). It is very different to say that something can serve as a substrate of change than to say that it is there BECAUSE it is an 'important mechanism'. The referee further objected to the postponing of any consideration of exaptation for sociolinguistic reasons until the end of the paper: socially relevant use of junk is important, and 'apparently pointless variation is socially functional'. First of all, not 'is' but 'may be'; it would be an extreme example of what I like to call the 'semiotic fallacy' to claim that everything in language is in some way 'meaningful'. My particular interest here happens to be with large-scale structural effects, not sociolinguistic ones, and that is a perfectly legitimate concern. But the objection as a whole, like the first one, misses a (chrono)logical point: the junk has to be there FIRST to be used at all, and my aim here is to clarify some of the ways in which junk is generated in the first place: in particular the role played by systemic collapse with non-functional residue. What happens to the residue later is, I maintain, open: if exaptation occurs it may be purely structural, with no social relevance, or it may be aimed at something socially indexical. There are no necessities involved, only possibilities.

IOO

EXAPTATION IN LANGUAGE EVOLUTION

REFERENCES

Anderson, J. M. & Ewen, C. J. (I987). Principles of dependency phonology. Cambridge: Cambridge University Press.

Antonsen, E. A. (1975). A concise grammar of the older runic inscriptions. Tubingen: Niemeyer. Bakker, R. T. (1975). Dinosaur renaissance. Scientific American 22. 58-78. Bendall, D. S. (ed.) (I982). Evolution from molecules to men. Cambridge: Cambridge University

Press. Bichakjian, B. (I984). L'evolution linguistique et la leqon de Lamarck. In Bots, H. & Kerkhof,

M. Misceldnea de estudios literarios, lingui'sticos e historicos oferecida a J. J. van den Besselaar. Amsterdam: Holland University Press. 245-249.

Bichakjian, B. (I987). The evolution of word order: a paedomorphic explanation. In Ramat et al., I987; pp. 87-108.

Bichakjian, B. (I989). Evolution in language. Ann Arbor: Karoma Press. Buck, C. D. (1933). Comparative grammar of Greek and Latin. Chicago: University of Chicago

Press. Cheshire, J. (I982). Variation in an English dialect. A sociolinguistic study. Cambridge:

Cambridge University Press. Collinder, B. (I960). Comparative grammar of the Uralic languages. Stockholm: Almqvist &

Wiksell. Dawkins, R. (I982). Universal Darwinism. In Bendall, I982; pp. 403-428. Dekeyser, X. (I980). The diachrony of gender systems in English and Dutch. In Fisiak, I980;

PP. 97-I 12. Dimmendaal, G. (I987). Drift and selective mechanisms in morphological change: the Eastern

Nilotic case. In Ramat et al. I987; pp. 193-210. Fisiak, J. (I980). Historical morphology. The Hague: Mouton. Gould, S. J. (1977). Ever since Darwin. Reflections in natural history. Harmondsworth: Penguin. Gould, S. J. (I983). Hen's teeth and horse's toes. Further reflections in natural history. New York:

Norton. Gould, S. J. & Vrba, E. S. (I982). Exaptation -a missing term in the science of form.

Paleobiology 8. 1.4-15. Hesse, M. (I966). Models and analogies in science. Notre Dame: Notre Dame University Press. Hoenigswald, H. & Wiener, L. F. (eds) (I987). Biological metaphor and cladistic classification:

an interdisciplinary approach. London: Frances Pinter. Jacob, F. (I982). Molecular tinkering in evolution. In Bendall, I982; pp. 131-144. Krahe, H. (I965). Germanische Sprachwissenschaft. II. Formenlehre. Berlin: de Gruyter. Lass, R. (i98oa). On explaining language change. Cambridge: Cambridge University Press. Lass, R. (198ob). Paradigm coherence and the conditioning of change: Yiddish 'schwa-deletion'

again. In Fisiak I980; pp. 251-272. Lass, R. (I984). Phonology. An introduction to basic concepts. Cambridge: Cambridge University

Press. Lass, R. (I986). Words without etyma: Germanic 'tooth'. In Kastovsky, D. & Szwedek, A.

(eds), Linguistics across historical and geographical boundaries. In honour of Jacek Fisiak on the occasion of his fiftieth birthday. Berlin: de Gruyter, I, 473-482.

Lass, R. (I987). The shape of English. structure and history. London: Dent. Lass, R. (Forthcoming). Of data and 'datives': Ruthwell Cross rodi again. Neuphilologische

Mitteilungen. Lass, R. & Anderson, J. M. (I975). Old English phonology. Cambridge: Cambridge University

Press. Lorenz, K. (1978). Vergleichende Verhaltensforschung. Grundlagen der Ethologie. Wien: Springer. Markey, T. L. (1976). A North Sea Germanic reader. Miinchen: Fink. Martinet, A. (1955). Economie des changements phone'tiques. Berne: Francke. Morpurgo Davies, A. (I987). 'Organic' and 'Organism' in Franz Bopp. In Hoenigswald &

Wiener, I987; pp. 8i-Io8. Mustanoja, T. F. (I960). A Middle English syntax. I, Parts of speech. Helsinki: Societe

N6ophilologique. Ostrom, J. H. (I979). Bird flight: how did it begin? American Scientist 67. 46-56.

IOI

ROGER LASS

Prokosch, F. (1938). A comparative Germanic grammar. Baltimore: Linguistic Society of America.

Raidt, E. H. (I980). Afrikaans en sy europese verlede. Van Tacitus tot van Wyk Louw. 2nd ed. Goodwood: NASOU Beperk.

Raidt, E. H. (I983). Einfuhrung in Geschichte und Struktur des Afrikaans. Darmstadt: Wissenschaftliche Buchgesellschaft.

Ramat, A. G., Carruba, O., Bernini, G. (eds) (I987). Papers from the 7th International Conference on Historical Linguistics. Amsterdam: John Benjamins.

Rijpma, E. & Scheuringa, F. G. (I969). Nederlands spraakkunst. 24th ed., rev. J. v. Bakel. Groningen: Wolters-Noordhoff.

Sanders, W. (1972). Oudnederlands. Tijdschrift voor Nederlandse Taal en Letterkunde 88. I6I-I77.

Scholtz, J. du P. (I98I). Taalhistoriese opstelle. Voorstudies tot 'n geskiednis van Afrikaans. Kaapstad: Tafelberg.

Sch6nfeld, M. (1932). Een oudnederlands zin uit de elfde eeuw. Tijdschrift voor Nederlandse Taal en Letterkunde 52. i-8.

Stevick, R. D. (I963). The biological model and historical linguistics. Lg 39. 159-69. van Loey, A. (197oa). Schonfelds historische grammatika van het Nederlands. 8th ed. Zutphen:

Thieme. van Loey, A. (1970 b). Altniederlandisch und Mittelniederlandisch. In Schmitt, L. E. (ed.),

Kurzer Grundriss der germanischen Philologie bis 1500. I, Sprachgeschichte. Berlin: de Gruyter. 253-287.

Wells, R. S. (I987). The life and growth of language: metaphors in biology and linguistics. In Hoenigswald & Wiener, I987; pp. 39-80.

102