You are on page 1of 47

Linguistic Change, Social Network and Speaker Innovation Author(s): James Milroy and Lesley Milroy Source: Journal

of Linguistics, Vol. 21, No. 2 (Sep., 1985), pp. 339-384 Published by: Cambridge University Press Stable URL: Accessed: 16/09/2010 11:23
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may 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 Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact

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

in GreatBritain J. Linguistics 2I (I985), 339-384.Printed

Linguistic change, social network and speaker

JAMES MILROY Department of Linguistics Universityof Sheffield


Department of Speech, Universityof Newcastle upon Tyne

(Received9 JanuaryI985)

Thispaperis concerned withthe socialmechanisms of linguistic change,and we beginby notingthe distinction drawnby Bynon(I977) betweentwo quite differentapproachesto the study of linguisticchange. The first and more idealized,associatedinitiallywith traditionalnineteenthcenturyhistorical linguistics,involves the study of successive'states of the language',states reconstructed by the applicationof comparativetechniquesto necessarily partial historicalrecords.Generalizations (in the form of laws) about the relationships betweenthese statesmay then be made,and morerecentlythe of 'possible'and'impossible'processes specification of changehas beenseen as an importanttheoretical goal. Thesecondapproach, associated withmodernquantitative sociolinguistics, involveslessidealization of the database.An important objective is to specify HOwlanguagespass from state A to state B in terms of both the social processesinvolvedand the effect on linguisticstructure of a given change. The major goal is to develop a theory which is sensitive both to the constrained and regularnatureof changeand to its relationship with social structure. Some sociolinguistshave borrowedquite heavily from older scholarsnotablyBailey,who has triedto applya 'wave' modelto contemporary data (Bailey, 1973). Similarly, Labovhas assessed,in the light of recentfindings, the theoretical approaches of nineteenth centuryhistorical linguistics; in one article,he compares'lexical diffusion'models of changewith those which claimthatchangecomesaboutas a resultof the operation of regular phonetic

on a previous draftof this paperby John We acknowledge withthankshelpfulcomments Versions of different PeterTrudgill andNigelVincent. DickHudson,BenRampton, Harris, in Liverpooland the Societas parts of it were read at the Sociolinguistics Symposium bothin September Linguistica Europaea meetingin Manchester, 1984.Helpfulcomments fortheBelfast research Financial werereceived fromparticipants at bothmeetings. support on which parts of the paperare based was receivedfrom the Social ScienceResearch Council(grantsHR3771,HR5777).The secondauthoralso receivedgenerousfinancial supportfromthe SimonFund,duringhertenureof a SeniorSimonResearch Fellowship of Manchester, at the University I982-3. This help is gratefully acknowledged.



laws. Conversely, the present can be used to explain the past, i.e. to shed light on historical linguistic problems (Labov, 1974a; Milroy & Harris, I980). What seems to be well established now is that variability of a structured and regular kind is characteristic of normal language use and is a key to understanding mechanisms of linguistic change. At the phonological level change appears to affect contextually defined subsets of phonological classes in a (generally) regular way, spreading through the community in waves in a manner controlled by extra-linguistic factors such as the age, sex, social status and geographical location of the speaker. Except where the ongoing change originates with a high-status group and is more or less consciously adopted by others, spontaneous speech appears to be affected earlier than the speech characteristic of more careful styles. These general principles have emerged fairly clearly from the work of Labov carried out over the last two decades (see Labov, 1972) and are confirmed elsewhere. To exemplify the quantitative approach, we comment briefly on one particular study. Eckert's (I980) account of Souletan, a dialect of Gascon, quite explicitly attempts to bring together the concerns of traditional historical linguistics and those of sociolinguistics. She examines the relationships between long-term phonological changes which affect whole classes of linguistic items, and the 'competence' of speakers who are involved in an ongoing linguistic change. Historical, geographical and synchronic variable data are analysed to illuminate the processes involved in an upward chain shift of the back vowel system of Souletan; the back chain shift is particularly advanced here and is still in progress. Eckert characterises the change in terms of waves, which affect one word class at a time; as we might now predict, items lagging behind in the shift occur in the speech of older informants, as stylistic variants. Thus, the item sulament 'only' occurs with [a] in careful speech, but with the more innovative [o] in rapid connected speech. Arguing that phonological rules (which reflect speaker competence) ought to be written in such a way as to reveal this pattern, Eckert gives a formal characterization of such a variable rule. If the language choices open to the individual are placed in this broader context, they may be seen as reflections of 'earlier' and 'later' overlapping states of a dynamic phonological system. The capacity of a variable rule formalism to handle linguistic constraints on the implementation of the rule may be seen as characterizing successive 'waves' of the change. Thus, it is argued, individual language behaviour is related to historical changes by rules which are seen as reflecting the competence of a speaker whose range of linguistic choices is congruent with the waves of change which proceed regularly through time and space. In a sense, Eckert's work, like much of our own, straddles the two approaches distinguished by Bynon, attempting to see how they fit together. Micro-level studies of this kind which are both 'sociolinguistic' and 340



'historical' appear to support the claims of Weinreich,Labov & Herzog

(I968) that linguistic innovationsmove systematically throughspace(social,

and historical)affectinglinguisticstructure geographical also in an ordered manner. The task of explaininglinguisticchange was, they argued, best dividedinto five main areas. These are, first, the very broad problemof UNIVERSAL CONSTRAINTS on
possible changes. As Weinreich et al. note (ioi) this is part of a larger

theoreticallinguisticissue and falls beyondthe scope of quantitative sociolinguistics. Within historical linguistics it has been examined by many scholarsincludingLass (I980) and Vennemann (I983). concerns the'intervening Second,theTRANSITIONproblem stageswhichcan be observed,or whichmustbe posited,betweenany two formsof a language definedfor a languagecommunity at different times' (Weinreich et al., I968: IOI). As we havealready a great noted,quantitative analysishas contributed deal here, showingclearlythat transitionis evidencedby variationbetween conservative and innovatoryforms,with the formergradually givingway to the latteras relativefrequency changes. TheEMBEDDINGproblemis concerned withdetermining in regular patterns both the linguisticand the extra-linguistic contextof change.Includedhere would be an accountof the phoneticenvironments most favouringchange and the relativerankingof theseenvironments. Much of Labov'sown work has addressed this issue (and see also Eckert,I980). Includedalso would be the vowel analyses of Labov, Yaeger and Steiner (1972) which follow Martinet's'chain shift' model. EXTRA-LINGUISTICaspectsof the embedding question maybetackled indirectly byinspecting thedistribution of innovatory linguistic formsin speechcommunities. Labovciteshis ownwork,in addition to that of Trudgill in Norwichand Cedergren in PanamaCity, as supporting the generalization that wheresoundchangein progressis located,linguistic variablesdisplaya curvilinear patternof distribution (often showingup on a graphas an unexpected 'crossover'pattern).Innovating groupsappearto be locatedcentrallyin the social hierarchy, and are characterized by Labov as upper-working or lower-middleclass. (Labov, I980: 254). Moreover, younger speakersuse more innovatoryforms than older speakers(both quantitatively and qualitatively) and again accordingto Labov's analysis, sexual differentiation of speechoften plays a major(but as yet not clearly understood) role in linguisticchange. The EVALUATION problem pertains principallyto social responses to change'at all levelsof awareness, fromovertdiscussion to reactionsthat are to introspection' quiteinaccessible (Labov,I982: 28). Thisembraces notions of prestige, attitudes to languages (bothovertandcovert),as wellas linguistic and notions of correctness. stereotyping The principal contribution of Labov himself to the EMBEDDING and EVALUATIONproblems has been,particularly in his New York City study,to providea GENERALmodelof the sociallocationof a linguistic innovationand


of the manner in which it spreads from a central point upwards and downwards througha speechcommunity.Shortlywe shall look at some of the problemsassociatedwith this model and indeedin much of this paper we shall be presenting a critiqueof partsof it. Until fairlyrecently,Labov had not attemptedto tacklethe fifth area of investigationoutlinedby Weinreichet al.: this is the ACTUATION problem, form: articulated herein its most challenging featuretake placein a particular Whydo changesin a structural language with the samefeature,or in the at a giventime,but not in otherlanguages as samelanguageat othertimes?This actuation problem may be regarded the very heartof the matter(Weinreich et al., I968: I02). It is the actuationproblemwhich we discussin this paper; appropriate it and somepartialanswers strategies aresuggested for addressing areoffered to the questions posedby Weinreich et al. Mostimportantly, we tryto explain have failed to make much headwayin tacklingthe issue why investigators in I968 as 'the very heartof the matter'. whichwas described Sucha programme as was presented et al. is not necessarily by Weinreich the best way of organizing a systematic studyof linguisticchange.Whilewe do not attempthere to offer a comprehensive critiqueof the paper, some difficulties shouldbe noted. First, although the five aspects of the problem of change have been presented as relatively discrete,they do in fact overlap.Whilethis does not in itself necessarily constitutea difficulty, a readingof Labov's I982 article whichreviewswork on the problemsup to that time shows clearlythat his interpretation of the issuescoveredby each of the five categories is different from Weinreich's (it was in fact Weinreich who was mainlyresponsible for the early sections of the I968 paper from which we have quoted in this section).So whilewe havedrawnfreelyon Labov'sreview,the categories are formulationwherethere appearsto be a discussedin termsof Weinreich's is compounded Thisdifficulty have discrepancy. by the fact thatwe ourselves sometimescategorizeda phenomenonratherdifferentlyfrom Labov. For example,whilehe regardsdiffusionof innovationas part of the TRANSITION it hereas an aspectof ACTUATION.Now it is certainly question,we havetreated clear that no single aspect of linguisticchangecan be discussedcoherently withoutreference to at least some of the othersspecified et al. by Weinreich But sincethe disagreements whichemergewhenattemptsaremadeto specify how phenomena shouldbe categorized are sometimesquiteradical,it seems reasonable to suggestthat the distinctionsdrawnin the I968 paperare not final and that they requirefurthercriticalconsideration if they are to serve as a comprehensive for the studyof linguisticchange. programme We look brieflynow at Labov'sattemptto tacklethe actuationproblem by first locatingthe innovatorsthemselvesand then examiningtheir social



characteristics and relationships within their own (Philadelphian) neighbourhoods2. His main conclusionsare as follows. (i) Speakers who lead sound changeare those with the higheststatusin theirlocal communities as measured by a social-class index. (2) Among personsof equal status 'the most advancedspeakers are the personswith the largestnumberof local contacts within the neighbourhood,yet who have at the same time the highestproportionof their acquaintances outside the neighbourhood'(I980: 26I). Labov thengoes on to comment'Thus we havea portraitof individuals with the highestlocal prestigewho are responsiveto a somewhatbroader form of prestigeat the next largerlevel of socialcommunication.' Both points are relevanthere.Labov presentsin effectone superordinate locus of change,viz. a centralpositionin the statushierarchy (and here his modelis implicitly one dependent on the existence of socialstratification) and one morerefined or micro-level locus,withina groupof roughlyequalstatus. The diffusionof changeis accomplished who have many ties by individuals withinthe close-knit and who also havea relatively community largenumber of outsidecontacts.Our own arguments, whicharriveat conclusionsrather differentfrom those of Labov, focus almost entirely on the position of linguisticinnovatorsin localisednetworkswhichare madeup of personsof roughlyequalstatus.We shallalso discussmorebroadlythe type of network structure associatedwith (often rapid)linguisticchangeand are less willing thanLabovevidently is to presenta modelbasedultimately on statusor class. Afterall, theseareno morethan rathercontroversial constructs (see Halsey, 1978, for an accessiblediscussion)and the universalapplicability of such constructsto theoriesof changeis dubious. We pass now to a discussionof what is meant by the term 'linguistic change', highlightingsome problems and ambiguities.Changes in the realization of two Ulstervowelsarethenreviewed in somedetail,to exemplify the principlethat evidenceof linguisticchangemay be found in data which are variableon historical,geographicaland social dimensions.Using the networkconceptdevelopedpreviouslyin this research programme (Milroy & Milroy, I978; L. Milroy, I980), the informalsocial ties of linguistically innovativegroups are examined,and a model of linguisticchange, based partlyon our own conclusionsand partlyon work by Granovetter (I973) is presented. Thismodel,whichsuggeststhat innovationsflow fromone group
study(Labov'sterm)is distinctfroma survey [21 A 'neighbourhood' in thatno effortis made to elicit comparable data from isolated individualssampledin accordance with some principleof random selection. Rather, the languageof speakersis investigatedwith attentionto theirpositionin relationto othersin theirlocal neighbourhoods. Thus,both andsocialstructure language in verymuchgreater maybe examined depth,at the expense of some socialand linguistic breadth.



to another through 'weak' networklinks is designedto offer a practical solutionto an aspectof the actuationproblem;as suchit is concernedwith SPEAKER innovation,of whichthe reflexin the languagesystemis a change whichis alwaysobserved postfactum. Finally,we suggest(morespeculatively) that the model is capable of elucidatingparticularproblemsof language changeand variationwhichso far have seemedquitemysterious.

Although the ultimate aims of historical linguistics may be to specify universals of change(whatis, or is not, a possiblechangeand, withinthe set of possiblechanges,the kindsof changethat aremoreor less PROBABLE), the methodologyof historicallinguisticshas ALWAYS been comparative. Nineteenth-century linguistics('comparative philology')aimedat RECONSTRUCTIONof proto-languages by comparison of sisterlanguages, and so the termCOMPARATIVE inlinguistics, acquired, anassociation withreconstruction3. Here, however,we use the term COMPARATIVEin a more generaland literal sense, without any necessaryimplicationthat reconstruction is aimed at. Thus, the comparisonof two attestedhistoricalstatesof the samelanguage is also a comparative method. also uses a comparative Sociolinguistics method,in that the languageof different individuals or groupsis compared. Thedifference is thatthe changes are observed,or arguedfor, at a micro-level ratherthan a macro-level. In a sociolinguistic analysis,the observation of changeis narrowed down to comparisons basedon age and sex of speaker, stylisticvariation and social grouping;observedsynchronicvariationcan be viewedas the counterpart ofchangeinthediachronic dimension. Inpractice, thesemicro-level synchronic areusuallysupplemented patterns by 'real-time'observations. Thetestimony of nineteenth-century and early twentieth-century observersof a speechcommunityare used to help to establishthe long-termdirectionof change (Labov, I972: I63-I7I). Notice that as soon as the methodology is extended to take accountof past statesof language,it becomesto that extentthe same as comparison of two or moreattestedpast states.Insteadof comparing two past states,we arecomparing a presentstatewitha past state.Theimportant differences thatdo existbetweensociolinguistics and historical linguistics (as describedby Bynon, 1977) depend fundamentally on the fact that sociolinguisticmethodis rooted in the present,whichmeans that thereis direct accessto therichdetailof variation in speech-communities. Thus,it is possible to observeveryfullyboth the linguistic andthe socialembedding of observed changes. It becomes possible to specify the constraintsimposed by a
[31 'In using the comparative method we contrast forms of two or more related languages to

determine their precise relationship. We indicate this relationship most simply by reconstructing the forms from which they developed' (Lehmann, I962: 83).




pre-existing of changewithinthat system languagesystemon the possibilities (Eckert, I980; Labov, Yaeger & Steiner, I972; J. Milroy, I976) and to suggestand evaluatethe possiblesocial motivationsof observedchanges. The questionof socialmotivationis not uncontroversial. On the one hand, thereis a view, pioneeredby Weinreich, Labov and Herzogand assumedin this paper,that the studyof socialmotivationsconstitutesan important part of any possibleexplanationof change.On the otherhand, thereis a strong tradition in language study of separatinglanguages from speakersand looking for some of the ultimateexplanationsfor change in languagesas
systems. As Lass (I980:

has put it:

Linguistshave, I would maintain,normallytreatedlanguageas if it were in fact an autonomousnaturalobject(or an autonomousformalsystem): 'languagechanges'- it is not (necessarily) speakersthat changeit... It is temptingto suggestthat the separationof languagesfrom speakers is partlya hangoverfrom the nineteenth-century insistenceon the 'life' of - a viewverycommonly the language of speakers independent expressed, e.g. by Trench(i 888: 224): 'For a languagehas a life as truly as a man or a tree. . . '. Although functional explanations(avoidance of homophony, etc., as discussed by Lass, I980: 64-9o) seemto takespeaker-strategies into account, theseandmostothercurrent explanations (e.g.physiological andpsychological ones) do not normallymake a PRIOR distinctionbetweenspeakerbehaviour on the one hand,and languageas a formalsystemon the other;theyaddress themselves to the explanationof changesobservedin languagesratherthan explanationof speaker-behaviour. Some of them are, in any case, seriously flawed(as Lasspoints out). What is clearis that functionalexplanations do not addressthe ACTUATIONPROBLEMas formulated by Weinreich et al. Such explanations mayaccountforsomeinstances of, e.g.avoidance ofhomophony, but theydo not explainwhyhomophonywasNOT avoidedin otherinstances. In general,theydo not explainwhy a particular changetook placeat a given time and in a given languageor dialect,but not in similarcircumstances in otherlanguages anddialectsor at othertimesin the samelanguage or dialect. If we are to addressthe actuationproblem(whichis 'the veryheartof the matter'),we must breakwith traditionand maintainthat it is not languages thatinnovate;it is speakers whoinnovate.Thereflexes of speaker-innovations are then observedin languagestates, wherethey appearas systematicand rule-governed linguisticchange. As the best-known findingsof sociolinguistics have tendedto concentrate on phonetic and phonologicalmatters,it has been possible for some to dismiss themas superficial, non-explanatory andpurely descriptive (Chomsky, 1975). Evenat the phonological level,however,theseapproaches havecalled into questionsomeof the theoretical positionsof the dominant(Chomskyan) paradigm. Partsof the phonological modelproposedin Thesound patternof


English (Chomsky & Halle, I968), if applied to variation in modern English phonology, necessarily distort and misrepresent the 'competence' of native speakers. More suitable phonologies have, in practice, to be constructed and many of the assumptions of SPE phonology are not borne out in such cases (J. Milroy, 1976; I98I). Furthermore, the findings of sociolinguistics are not confined, as is often suggested, to the description of configurations of surface PHONETIC variants: it is in sociolinguistic work and not, as far as we know, in other approaches, that we can observe cases of rule change in progress (and therefore hope to explain such phenomena). We now briefly discuss an example: the gradual loss of /a/ raising after velars in Belfast. David Patterson (i86o) attests that /a/ was raised to [E] after the velar consonants /k, g/, and lists given by Gregg (I964) for the Ulster Scots town of Larne suggest that this rule applied regardless of FOLLOWINGconsonant. In present-day Belfast we have attested no cases of raising after /g/, and the rule is variable after /k/. It is variable to the extent that male working class speakers can vary between [c]and [a, a] in the same lexical items. Their choice reflects the application of different rules, rather than application or nonapplication of a single rule. Thus TM (Clonard) has [kig n]: 'can', rapidly followed by three tokens of [kia n]: 'can', in succeedingutterances.As we shall see in Section 3, the trend in the /a/ system has been towards backing of /a/ since Patterson's day (conditioned by the following consonant), and in these post-velar environments the rules are in conflict. The choices open to speakers in monosyllables may be listed as.instructions, as follows: (i) After /k/, choose either mid or low, unless /r/ follows, in which case low must be selected, (2) If mid is chosen, select low-mid, short [a] before voiceless stop; otherwise select mid, long, (3) If low is chosen, select short front [a] before voiceless stop; otherwise select long, back [a ].
. .

Clearly the rule for raising after velars is recessive: it has disappeared after /g/ and before /r/ (Patterson has care to represent the pronunciation of 'car'), and is otherwise variable for many speakers. For many younger East Belfast speakers, however, it has actually disappeared. In general, it is the following consonant more than the preceding one that dictates which realizations of /a/ are adopted. We shall see in Section 3 that the choice listed in 3 (above) is somewhat idealized: in fact there are greater and lesser probabilities of backing in an ordered series (depending on following consonant), and some environments are more likely than others to allow back-raising and rounding to [3]. (For other examples of rule-change in progress, restructuringand merger, see J. Milroy, i984b; J. Milroy & Harris,

In presenting such a configuration of change it is clear that we are primarily




describinga state of languageratherthan the 'competence'of individual speakers.The speakersthemselvesmay, or may not, have access to all the possible variants,and (as shown in J. Milroy, i982a), some middle-class exhibitlittlevariation; speakers individuals mayconvergefor all itemsEITHER on front[a]orback[a]. Suchspeakers maybe said- in a linguistically-oriented - to have 'lost' the rules for raising and backing in different dimension environments.In a speaker-oriented dimension,however, these speakers merelydisplaya different pattern,and we cannotassumethat they everhad the rules for frontingand backingin their active competence.Nor do we know whetherthey are awareof themin theirpassivecompetence. In other cases, speakersmay be observedto vary in their realizationsof the same lexical items in the same phonologicalenvironments; such speakershave variablerules. Thus, when we consider speaker-competence, there are difficultiesin specifying what a linguisticchangeactuallyis and how it is implemented. At the macro-level, claims for changehave normallybeen assumedto rest on an observeddifference betweenStateA and StateB, and have not depended on speaker intuitionor competence for instance, (twentieth-century speakers, are not assumedto have intuitionsabout fourteenth-century states).At the micro-level,in which observed change depends on variation in speechcommunities,speakerintuition has been assumedto be relevant,in that speakersmay have access to both recessiveand incoming variants and know when to use them. Even at this level, however, it seems that speaker-behaviour varies,andit is possiblethatindividual speakers have- to a degree- differentialcompetence and intuitions. The difficultyis that linguistic change must presumablyoriginate in speakers rather than in languages.We thereforefind it convenientto distinguish betweenlinguistic CHANGE, on the one hand, and speakerINNOVATION on the other. It is the originand diffusionof SPEAKERINNOVATIONSwith which we are concerned in this paper. Speakerinnovations,like otherinnovations,may be classified in termsof theirsuccessin subsequent diffusion,as follows:

A speakerinnovationmay fail to diffusebeyondthe speaker. A speaker innovation maydiffuseinto a community withwhichhe/she has contact,and go no further. (3) A speaker innovation maydiffuseinto a community withwhichhe/she has contact and then subsequently diffusefrom that communityinto othercommunities via a furtherinnovatorwho has ties with both the relevantcommunities. When the resultsof this processare observed, we tend to label the resultsas 'linguisticchange'.The set of possible communities throughwhich such a changecan diffuseis in principle infinite,and althoughlinguisticand socialconstraints on a changecan



in some instances be specified, the limits of POSSIBLEdiffusion cannot be precisely stated - either in terms of space and time or in terms of the possible states of language or society that may favour or disfavour the change. It is not suggested in the present state of our knowledge, that the innovators can be precisely located. The linguistic innovator to whom we refer is as much an idealization as Chomsky's 'native speaker-listener', and it is our aim to model the sources and processes of linguistic innovation in more detail than has been possible in the past. We consider arguments about probabilistic grammars and the status of variable rules (Romaine, I98 I) to be, in principle, irrelevant here. For, although much of the data presented in this paper has been collected from speakers and (necessarily)subjectedto quantification, our arguments are not based on quantities, but on processes that have been observed to take place in speech communities. Although such processes may have been analysed quantitatively, they are not in themselves quantitative phenomena. By using such methods, however, we may have made some progress in locating the idealized speaker-innovator. We end this section by commenting on (I)-(3) above (pp. 347-348). Notice that speaker innovation is not identical with linguistic change. As (I) implies, some innovations may not be accepted by a community and hence may not lead to change. On the other hand, speaker innovation may lead to a change in one segment or part of the grammar,which then sparks off a chain reaction that seems to be internal to the language system. Thus, in the English Great Vowel Shift, it may be argued that ME a was first raised, and that as a consequence of this, the ME vowels above it in phonetic space were also raised (or diphthongized in the case of the highest vowel). In such a case, it is possible that speaker innovation is relevant only to one vowel and that movements in the other vowels are motivated by the language system. Yet even here it must be admitted that speakers have been motivated to keep vowels distinct within the system. With referenceto (3) above, we must also note that, as the limits of possible speech communities (like the limits of social networks) cannot be specified, we do not know that a change observed to have entered a community (through the activities of certain speakers or groups) is in fact original to those who are observed to carry the innovation. The apparent innovation may already have been well established in some other community, and this in turn may have adopted the innovation from elsewhere. In observing change in a given community, therefore, we do not know beforehand at what point in a cycle of change we have entered the community. Although, from a synchronic point of view, certain individuals and groups may be identified as innovatory (see Section 3 below) and as responsible for introducing an innovation to their immediate communities, it is possible that the change concerned has had a long history elsewhere. We shall see that this is so in 348



the case of changesin the vowels /E/ and /a/ in Belfast- to whichwe now turn. 3.




If we compare the range of variation in Belfast vernacularvowels with of RPit is obviousthatmanyof themhavea startlingly text-bookdescriptions (J.Milroy,1976; I98I; I982). Realizations rangeof realizations widephonetic of /a/ rangefrom [?] or above beforevelarconsonants,as in bag, bang, etc, to /c/, in hand, bad, etc. This is furthercomplicatedby a variation in tend to vowelsin monosyllables Briefly, and diphthongization. vowel-length of consisting and before clusters stops voiceless be short before voiced conand before fricatives + voiceless stop; they are long sonorant sonants.Closingdiphthongs([a i]) can also occurbeforevoiced velars,and centringdiphthongs([a-a][a-e])occur when the vowel is back, long and, especially, also raised and rounded. The range from front to back is in Table I. represented
[?-] [m]



bag bang
Front only Back only Front back

bad bad bat grass grass snap hand hand ant snap back Velarenvironments Fricative& voiced consonant environments(excluding velars) Voiceless stop environments(excluding velars); back variantsattestedonly amongstEast Belfastyouths back
Table i

of phoneticrangeof /a/ in representation Simplified using key words Belfastvernacular, the range The range for /e/, in e.g. step, bed is also wide. Qualitatively, apply, with centring is from [a] to around[e]: similarrules of vowel-length (see diphthongsof the type [q-a]tendingto develop in long environments Table2). Some Sucha widerangein two neighbouring vowelsresultsin overlapping. realizationsof /E/ are like realizationsof /a/, and vice versa. Projecting has takenplace in time, it is possibleto arguethat restructuring backwards 349


[e:,,e -a, ?'-, ? * ]

Low [a, w] wet, went (Voicelessstop, sonorant+ voicelessstop environments)

Table 2

bed, bend, best (Fricativeand voiced consonantenvironments)

Simplified of phoneticrangeof /?/ in Belfastvernacular, representation usingkey words

at some time in the past and that, for example, the /a/-/c/ distinction may

have been neutralised beforevelars,with bag and beg havingbeen possibly identical.However,restructuring (with transferfrom /a/ to /?/ and vice from the present-day versa) cannot be adequatelydemonstrated evidence. in disentangling Althoughsomespeakers havedifficulty pairslikepack/peck, areawarein most casesthat [E] speakers realizations beforevelarsare tokens of /a/, whereas[a] realizations beforevoicelessstops are tokens of /?/. As raisingappliesto /a/ beforevoicelessand voicedvelars,itemslike back,bag areoftenrealized with [c]. Howeversincelow realizations of /E/ applybefore ALL voicelessstops (including the velar),itemslike neck,wreck(withvelars) tend to be realizedwith [a]. This resultsin an apparentflip-flop,and the followingexamplesare typical: 'The back [bEk] of my neck [nak]; 'Will you pay by Access [EksEs] card or by cheque[tfak]'; 'Jet [tat] - lag [E 9]'. Thereare two overlapping systems,informallystatedas follows: Velar /a/-[E]/VoicelessStop /?/+ [a]/The complexity of such systems, together with the range of socially motivatedvariationthat occursin the realizations of the vowels, presenta considerable challengeto our abilitiesto identifythe directionof changein progress,but the sheeramountof variationprovidesmany clues. First, we considerthe regionaland social rangeof realizations of /?/. (i) Raisingof /?/ FigureI showsthe resultof a quantitative analysisof /?/ realizations in two Belfastouter-citycommunities (Andersonstown and Braniel)and a smaller town(Lurgan) situatedI 7 milessouth-west of Belfast.ThesymbolT indicates a followingvoicelessstop or sonorant+voicelessstop cluster;C$ indicates that the vowel is in the stressed syllable of a polysyllabic word (this

LINGUISTIC 70 60 50 40 30 2010 _ 0 TC$D



ANDERSTOWN (n = 1104)






70 60 50 _ 40 30 20 10 0

BRANIEL (n= 800)







7060 5040 30 20 10 0

LURGAN (n=1484)







Figure i Percentage distribution of /E/ (bed, bet) variants by following environment in outercity Belfast (Andersonstown, the Braniel) and Lurgan. (After Harris, 1983: 157.)



environment tends to favour short realizations); D indicates following fricative or voiced consonant (excluding /r/). Notice that the lowest short realization, [a], is not favoured, but that in Lurgan short and low realizations in short environments (T, C$) are more favoured than elsewhere (see also below), and that long realizations [ae:,?'] in these short environments are rarer in Lurgan. The inner-city figures (Ballymacarrett, Clonard, Hammer) in Table 3 clearly show some contrasts with the outer-city figures. Before

Men 40-55 Women40-55 Men 18-25 WomenI8-25 T B

C H C$ B C H

8I 75 56 67 68


73 67 50 6o

97 97 73 8i 76

84 98 78 75 76

Table 3 Percentage low realizations of /?/ in typically 'short' phonetic contexts in three inner-city Belfast communities, Ballymacarrett(B), the Clonard (C) and the Hammer (H)

voiceless stops, a low short realization ([a], [2]) is categorical for many male speakers, while the women more often prefer higher and often lengthened realizations. Thus, for typically low vowel environments, as in wet, went females often have [w?:t, w*-:nt] for 'vernacular' [wat, want]. In this respect the inner-city female pattern is similar to that found generally in these higher status outer-city communities. These variable data give us a basis for examining processes of change, since they suggest initially that either the higher or lower variants are innovatory, or - more properly - that the direction of change is either raising or lowering
of /?/.

In fact, an examination of historical documentation (real-time evidence) suggests that the direction of change is towards raising. Moreover, it appears that mid realizations are gradually appearing in environments (such as pre-voiceless stop) where low realizations were once the norm. It also appears that as the low variants are replaced by higher ones, the relevant vowels are lengthened and sometimes diphthongized: thus, as the rules are applied, conservative variants such as [rant, rent]: 'rent', are replaced by [rE nt] (raising and lengthening) and [re.ant] (diphthongization). (For a discussion see J. Milroy, 1976). The options open to speakers for the realization of /c/




beforeVoicelessStopor beforeConsonant + VoicelessStopmaybe described as follows:

(i) (2)

Choose eithermid or low; If low, realizeas short; (3) If mid,realizeas long; realizeas monophthongor diphthong. (4) If mid-long,

This is of coursean idealizedand simplified account,and the aim of listing suchoptionsis descriptive only. We do not claimthat we know the ordering of rules,and if we do wish to orderthem,it is possiblethat LENGTH should precede HEIGHTor thatlengthening andraisingaresimultaneous. Nor is there any implied claim that all individual speakers have the same rules or - far from it. For the great complexitiesthat do exist when rule-order
speaker-variation is studied, see now Milroy et al. (I983), Harris (I983).

Acceptingthis as a broaddescription of the currentstate, we now examine some real-timedata in orderto confirmthe directionof change. Pattersongivesa list of fivewordsof the /s/ class,whichwerethen(i860)
pronounced in Belfast with low realizations: wren, wrestle, wretch,grenadier,

desk.These few examplesare enough to show that the low realizationwas thenmorewidespread thantoday:wrenand deskdo not satisfythe Voiceless Stop or Sonorant +Voiceless Stop conditionin monosyllables, and are now categorical[e:]or [j -a] environments. Even the disyllablewrestle is unlikely to appearwith [a], as the rule for raisingand lengthening before [-s] now almostalwaysoverrides the tendencyto lowerand shortenin disyllables and polysyllables.4 Items like wretchand grenadierare now variable. Staples (I898) and Williams(I903), additionally give quite detaileddescriptions of the vowel in the city, whichallow us to infer that low variantshad a much widerdistributionthen than they do today. The completelist, taken from thoseearlywriters allowsusto seethatthelowvowelappeared in environments whereit wouldnot appearnow - for example,beforevoicelessfricatives and voicedstops(HarrisI983: I60). Thedistribution in present dayBelfastis quite different, as is shownby Table3 and FigureI. In conservative workingclass speech,low variantsare maintained in 'short' environments, very much as in the nineteenthcentury:but low realizationshave been almost entirely replacedin long environments by mid realizations of /c/. More prestigious and less conservative speakersare less likelyto use 'low' realizations, even in shortenvironments. It is evidentthat over the last hundredyears or so mid realizations have been spreading at the expenseof low realizations. Mid /?/ has now almost totally replacedlow /E/ in 'long' contexts (pre-voiceless stop, pre-sonorant+voiceless stop, and in polysyllables).Low status inner-cityspeakers
[41Itemslike wren,wretch, wrestle werehistorically affected by loweringafter/w/, and still varieties with a low vowel. appearin manyIrishand American



low realizations in short environ(males)sometimesstill have categorically ments,butin themoreprogressive outer-city housingestates,thevowelis now midfor somespeakers. categorically thedistribution of variants Interestingly, in Lurganis moresimilarto that of the inner-city areasthanthat of the outer areas (a patternthat appliesalso to other vowel and consonantvariables). This relatively rapidlinguisticchangein Belfasthas accompanied its rise in populationfrom about 120,000 in I 860 to nearlyhalf a millionin the early yearsof this century,and Belfastmay be takenas an exemplarof linguistic changein fast-growing communities (whileruraltownsandvillagesadhereto older patterns).The characteristic NETWORK structuresof these different types of communityare also relevantto the mannerin which changemay come about, in so far as urban growth tends at first to weaken strong pre-existing ruralnetworks. Wemay supplement our observations on /?/ by considering evidencefrom present-day Ulsterdialects.Theseare dividedinto two distincttypes.Ulster Scots dialects are found in East Ulster in a belt extendingfrom around Coleraine in theNorth,through most of CountyAntrimandmuchof County Down (whichis south of Belfast- see map). Most of Ulster to the west of this belt is English-based or mixedScots-English. Present-day Belfastdialect is often describedas an intrusionof this Mid-Ulstertype into the Scottish belt.Now, thelongmidvariants eastern of /?/ areoverwhelmingly associated withpresent-day UlsterScotsdialects(Gregg,I972) and arecharacteristic of moderncentral Scots dialectsgenerally(an exceptionis very conservative
Galloway Scots, on which see J. Milroy, I982b). Traditional Mid-Ulster English, on the other hand, is characterized by lower realizations in all environments (Harris, I983: I8I). The pattern of distribution in these dialects is remarkably similar to that of nineteenth-century Belfast vernacular as described in Patterson, Staples and Williams. We may infer that this pattern is a residue of some earlier English vowel pattern that has not been well identified or described by historical linguists. There is sixteenth-century orthographic evidence (discussed by J. Milroy, I984b) that suggests some distribution of low vowel realizations for /?/ in London English of the period: it seems possible that this pattern of lowering of historic short vowels has been overtaken in recent Standard English and Central Scots by a pattern of raising and (in the latter case) lengthening. The Mid-Ulster dialects may therefore have preserved to a great extent an older general English vowel pattern, and they may help us to project knowledge of the present on to the past. The historical and geographical evidence then both suggest that the low realizations of /E/ (conservative English in background) are giving way in a linguistically ordered way to the long mid realizations characteristic of present-day Scots. It is clear that this change carriesprestige in Belfast in terms of social class hierarchy and status, as it is the more prestigious groups that tend to adopt it and the more 'advanced' (generally female and younger) group who introduce it to the conservative inner-city communities (which are 354




Rathlin Island






Map representing'core' Ulster Scots areas of north-east Ulster (shaded areas): adapted from Gregg (1972).

characterizedby dense and multiplex network ties that tend to resist innovation andmaintain conservative innovative forms).Thetensionbetween and conservativesocial mechanismsgives rise to a identifiablepatternof whichmaybe represented as a historical shiftfroman older gradual diffusion, of modernScots.As we English-type patterntowardsa patterncharacteristic haveimplied,themanner in whichthechangeproceeds is conditioned by both of changein social and phonologicalfactors.We now turnto a description /a/, with whichthe /e/ systemcan be compared.
(ii) Backing of /a/ As we have indicatedabove (p. 349), the range of realizationsof /a/ in - from [e]through[a]and [a] Belfastvernacular is considerable present-day 355


to back raised and rounded [3]. Again, as for /?/, patterns of lengthening and diphthongization are present, with long vowels being associated mainly with back realizations and with the higher front realizations before voiced velars (see Table I). In what follows, we are concerned only with backing and retraction, and we therefore largely exclude the pre-velar environments (in which backing is not found). Table I also shows that back realizations are favoured by following fricatives, non-velar voiced stops and non-velar nasals (on the rules for preceding velars, as in cab, carrot, see p. 346 above). Nasals favour backing particularly strongly. Middle-class urban speakers (J. Milroy, i982 a) tend to narrow the extreme range described above and in some cases converge on a point somewhere in the middle of the range, around [a] (but see below). The widest range is found mainly in the speech of inner-city male speakers. Furthermore, it is the MALES of Ballymacarrett (East Belfast) who use the backed variants most and who show evidence of spreading the backed realizations into VOICELESS stop environments (as in that, wrap), where short, front variants are expected. If there is evidence of change in progress towards backed variants of /a/, it will therefore be male speakers who are leading it, rather than the females who lead the change towards raised /?/. Historical documentation suggests that /a/ backing is a recent trend. The elocutionist Patterson (i86o) does not comment on /a/ backing at all. On the contrary, his remarks suggest that the Ulster tendency was towards fronting and raising and that the most salient Belfast feature was fronting and raising in velar environments. In some places [presumably in the north of Ireland: JM, LM] the short sound of e is improperly substituted for a, in almost every word in which it occurs; in Belfast, however, this error is almost exclusively confined to those words in which a is preceded by c or g, or followed by the sound of k, hard g or ng. (Patterson, i86o: I5) A very few of Patterson's spellings may indicate that /a/ backing and rounding had been observed sporadically in -r and -l environments: he has form for 'farm' and canaul for 'canal'. However, examples of this kind are so few that they indicate only a slight tendency (possibly confined to some pre-sonorant environments), which is not enough for /a/ backing to be discussed as a stereotype. The item car appears in Patterson as 'care', in which the now highly recessiverule for fronting and raising after velars is clear. Items like hand, band, in which [o] is now stereotypically expected, are given simply as han, ban, etc. Frequently, however, items that now have low and/or back vowels, are given with [E]: these include rether for 'rather' (a rural Scots residue), e for a in single nasal environments in polysyllables such as exemine, Jenuary and in nasal cluster environments such as demsel, exemple, Entrim ('Antrim'), slent, bendy '(bandy'), brench. 356



inclinedtowards accountindicatesa systemgenerally Whereas Patterson's front-vowel realizations, Staples(I898), writingnearly40 yearslater,reports a 'low backwide'vowelbeforenon-velar nasals,in e.g. man,hand,land.Since Patterson's time -/r/ environments have become categorically back realizations.Otherwise,the figureson present-dayvariationconfirmthat led the change, that have subsequently since then it is nasal environments In East Belfast, closely followedby fricativeand voiced stop environments. as we have noted above, backing is spreadingeven into voiceless stop and this is most clearlyattestedin young men (those in our environments, samplewereaged I8-20). THUS, ALTHOUGH RAISING AND LENGTHENINGOF f/? AND BACKING OF /a/

clear from patternsof stylistic variationthat (as we might alreadyhave the two changeshavedifferent inferred) prestigevaluesattachedto them.As Table4 indicates, thebackingof /a/ tendsto be resisted in careful by speakers East Belfast(Ballymacarrett) Women(40-55) Men (I8-25) 3.43 West Belfast(Clonard) Women(4-55) Men (I8-25)
1.77 I.85 Table 4 2.36 2.33 I.75 2.58 2.89

Men (40-55) IS SS
3.03 3.58


Men (40-55) IS
SS 2.79 2.79

2.36 2.6I

Incidenceof retractionand backingof /a/ by age, sex and conversational calculated style in two Belfastcommunities, by an indexscorerangingfrom o (minimum) to 4 (maximum). IS, interviewstyle; SS, spontaneousstyle 'interview'style(whereas raisingof /F/ is MORElikelyin carefulstyles).Thus, men seem to be principally associatedwith a change that speakersdo not consciouslyview as beingof high prestige,whilewomenare associatedwith one adopted by speakers in their more carefully monitored styles (for furtherdiscussionof these figures,see Section6 below). Ourreal time evidenceconfirmsthat the movementin /a/ is phonetically from frontto back.This meansthat sporadicfront-raising (foundmainlyin
West Belfast) in words likeflat, trap ([flet, trep]) must be seen as residues and

not as innovations.The belief of many casual observersthat raisingbefore velars(and very occasionallybeforevoicelessnon-velarstops) are attempts


RP ('ReceivedPronunciation', as described, to imitateconservative e.g. by is also shown Gimson, I980) to be wrong by quantitativeand diachronic evidence.The quantitative evidenceshowsthat the frontingand raisingrule in Belfastvernacular is virtuallyconfinedto velarenvironments and cannot applyto wordslike bad,hand,stab (whichare front in RP). The diachronic evidence shows that, for a century or more, the trend has been towards retraction and backing. Theevidence also indicates thatthe rulefor backingdiffuses geographically from East to West Belfast(see Table 4). Scoresfor /a/ backingare higher for East Belfastmales than for any other groupsstudied,and the rangeof environmentsin which backing operates is extended to voiceless stops amongst younger East Belfast males. It appearsto be inner East Belfast (Ballymacarrett) that providesthe modelfor working-class speechin the city (L. Milroy, I980); this is discussedby Harris(I983) in termsof a 'labour aristocracy' representedby the (relatively) fully employed protestant populationof East Belfast. Both /a/ backing and /?/ raising are relativelyrecent phenomenain Belfast(but see below),and both are associatedwith a background in Scots. Patterson'saccount of Belfast shows characteristics of conservativerural Scots lexicaldistribution, much of whichappearsto have been residualand is now obliterated by restructuring. Howeverlengthening and raisingof /e/ andbackingof /a/ aremodernScots.Gregg's(I972) accountof UlsterScots givesoverwhelmingly backrealizations of /a/ anddescribes /c/ as oftenlong in realization (contrastthe veryshortlow realizations in conservative Belfast vernacular, such as [stap,t3at]for step,jet). Similarly, /a/ backingseemsto be a verygeneralmodernScots feature(Lass, 1976).East Belfastadjoinsthe Ulster-Scots regionof North Down (wherebackingis strong),whereas West Belfast points south-westdown the Lagan Valley, the speech of which is Mid-Ulster with less Scots influence;furthermore, immigrationto West Belfast is recent and is largely from a Mid- and West-Ulsternon-Scots hinterland. Presentday quantitative studiesin Lurgan,a smallcountrytown south-westof Belfastin the Lagan Valley,confirmthe existenceof an /a/ systemwith little backing(front vowels have been noted in that area even before[r]and finally),whichis quite similarto Patterson'si86o accountof Belfastin this respect(Pitts, I982). Finally,we mustnote that if we take a generaloverview,thesetwo vowels appearto be movingaway from one anotherin phoneticspace,ratherthan in the samedirection(as we would expect,e.g. in a chain-shift). We are not in thispaperprimarily concerned withthe embedding of changesin language systems(and arguments based on this could suggestthat one change- /-/ raising- is slightly more recent in origin than the other), but we may commentthat if data for individualspeakersand homogeneousgroupsare the overallpictureof vowelsmovingawayfromone anotherdoes examined, not appearso prominently as it does when we focus on the languagerather 358



than the speaker.Speakerswho have [6] raisingtend to have more fronted realizations of /a/ (thesespeakers areoftenfemale),andthosewhohavelower reali"ations of /?/ aremorelikelyto have[a]backing(thesespeakers areoften male). Thus, an account based on what speakersactually do looks very different from a generalized accountof changein the languagesystem. In the next section,we move from an accountof the languagesystemto a discussionof speakersand theirsocial role in phonologicalinnovations. 4. SOCIAL NETWORK STRUCTURE ANALYSIS OF /a/ AND /6/

In additionto the variablesof age, sex and statusdiscussedin Section3, a witha speaker's DEGREE OF INTEGRATION into further associated socialvariable his closeknit communityappearedto affect the probabilityof his being innovativewith respectto choice of vowel variants. linguistically ties Generallyspeaking,it seemsto be truethat the closerthe individual's to a local communitynetwork, the more likely he is to approximateto vernacular norms (see L. Milroy, I980, for details). Following some welldeveloped anthropologicalfindings, we have suggested that a closeknit network has an intrinsic capacity to function as a norm-enforcement mechanism,to the extent that it operates in opposition to larger scale institutionalstandardising pressures.One corollary of this capacity of a closeknitnetworkto maintain normsof a non-standard kindis that linguistic will be associatedwith linguistic the LOOSENINGof such a networkstructure
change (L. Milroy, I980: I85; Gal, 1979). It is the implications of that

here. corollaryon whichwe concentrate A major point emergingfrom our earlieranalysisof language/network relationshipswas that the variableNETWORK needed to be consideredin relationto the variableSEX OF SPEAKER. Indeed,as Gumperzhas remarked (I982: 71), the networkvariableis in generalclosely associatedwith many others,including generation cohort,geographical location,and socialstatus. Thus,ournexttaskhereis to pickout briefly therelevant partsof our analysis of the social distributionof innovatory realizationsof /a/ and /?/, as identified in Section3. Firstof all, realizations of /a/ and /?/ arestronglyaffected by the variable SEX OF SPEAKER.Thus, althoughincomingvariantsof both vowelsappearto have originatedin the same hinterlandScots dialect, each has assumeda diametrically opposedSOCIALvalue in its new urbansetting. Raised variants of /6/ are, in the low status inner city, associated particularlywith women and with careful speech styles. They are also associatedgenerallywith slightlymore prestigiousOUTER city speech,and data collectedby surveymethodsconfirmsthat the higherthe statusof the speaker,the morelikelyhe is to use raisedvariants(see Milroyet al., I983). Differentlevels of use accordingto SEX OF SPEAKERare particularly evident


in Ballymacarrett, where it appears to be YOUNGER FEMALE speakers who are most strongly associated with the incoming raised variants. The incoming variants of /a/ show an almost perfectly converse pattern of social distribution. High levels of backing are associated with males (particularly Ballymacarrett males, although levels in other inner city areas are still quite high) and with casual styles appropriate to interaction between peers. The most extremely backed variants do not appear at all in outer city speech. Interestingly, the sex differentiation pattern across the three inner city areas is not as consistent for /a/ as it is for /?/; there is some indication that the young Clonard WOMENare increasing their use of backed realizations when compared with other female groups (see Table 4). They also use these variants MORE than their male counterparts although they follow the expected sex differentiation patterns with respect to other phonological variables (see Section 6 below for a discussion of the Clonard pattern). In summary then, it appears that incoming variants of /a/ are associated with core Belfast vernacular, while incoming variants of /?/ are associated with careful higher status speech. If we look at the relationship between speaker choice of variant and individual network structure, the picture becomes even more complicated. With respect to both vowels, choice of variant shows a correlation with personal network structurein some subsections of the inner city communities; but the details of this correlation are quite different for each vowel. The vowel /a/ is particularly sensitive to variation according to the network structure of the speaker; but WOMEN appear to correlate their choice of variant more closely with their personal network structure than do men. This means that among women a relatively large amount of /a/ backing is more likely to be associated with a high level of integration into the network than is the case among men - a relationship analysed by Spearman's Rank Order Correlation (L. Milroy, I980: I55). Although, as we have noted, women are much less likely than men to select back variants of /a/, this generally lower level of use does not prevent individual women from varying their realization of /a/, within the female norms, according to their social network structure. Thus, the DEGREEOF FIT between phonological choice and network structure may be seen as an issue quite separate from the ABSOLUTE LEVELOF USE of a particular range of variants. We may thus argue that /a/

functionsfor womenas a NETWORK MARKERto a greaterextentthan it does

for men; by this we mean that there is for them a higher correlation between choice of variant and network structure,a tendency to select relatively backed variants being associated with higher levels of integration into the community. When we look at the relationship between choice of /?/ realization and individual social network structure, we find a pattern emerging converse to the one described for /a/; recall also that the incoming variants of the two 36o



with regardto status, vowels showedan almostconversesocial distribution sex of speaker,and speechstyles. Most importantly, thereappearsto be no tendency at all for womento use /s/ as a NETWORK MARKER in the sense describedabove; but there is a significant correlation betweennetworkscoresof MALEspeakers (particularly young male speakers)and choice of /f/ realization.A tendencyto select relatively LOW (conservative) variants is associated witha relatively highlevel of integration into the community(see L Milroy, I980: 155 for details). This complexrelationship betweennetworkstructure, sex of speakerand languageuse is summarised in Table 5. However,our interesthere is in a with High correlation networkstrength Females
Males Table 5

Changeled by /a/


Contrastingpatterns of distributionof two vowels involved in change, accordingto sex of speaker,relativefrequencyof innovatoryvariantsand with networkstrength level of correlation whichwe are now able to make concerning on the one hand generalization the relationship and on the other betweenlanguageand networkstructure, the socialidentityof the innovatinggroup.IN THE CASEOF BOTH /E/ AND /a/

of this language/network that fulfilsa cohesive relationship (a relationship socialfunction) socialgroupto adopttheroleof linguistic enablesa particular innovators.This appearsto be the case regardless of whetherthe innovation is evaluatedby the widerurbancommunity as beingof high or of low status. For althoughit is clearthat /e/ raisingis diffusing on a muchbroadersocial front than /a/ backing,the generalization still seems to hold true that it is those personsin the innercity for whom the vowel functionsless clearlyas a network marker who are the principal innovators into their own communities. It is important to note thateventhoughbackedvariants of /a/ arestrongly emblematic of vernacular are nevertheless speech,they to higher spreading statusgroupsin thewidercommunity. Butthisdiffusion is beingimplemented
in a manner very different from that affecting /?/. We have noted that

raisingis characteristic both of low-statusfemalespeechand moregenerally



LIN 21


of higherstatusspeech.The diffusionof [E] raisingon this wide social front is confirmed both by linguisticsurveydata and by more detailedouter-city communitystudies. Whenwe look at the social distribution of variantsof /a/ (on which see J. Milroy,i982 a) wefindby wayof contrast thathigherstatusBelfastspeakers avoid both extremefront AND extremeback realizations,as they converge aroundCardinalVowel 4 in the middleof the phoneticrange.However,a veryinteresting groupof young,male,middle-class can be identified speakers in the sample of speakers studied in the survey. They also show the characteristic middle-class tendencyto convergearounda limitedphonetic area,with relatively little conditionedvariation.However,phonetically, the point at whichtheyconvergeis further back thanthat characteristic of older middle-class speakers. It appearstherefore that the mechanism of diffusionassociatedwith each of t;he vowelsis different. Raisedvariantsof /?/ are apparently in spreading a linguistically ordered way,with'long' environments affected first.Formany outer-cityand middle-class in speakers,a raisedvowel is alreadycategorical all environments. Althoughbackedvariantsof /a/ appearto be diffusing historically and laterally (throughthe low statusinner-city in a communities) linguisticallyorderedmanner parallel to the processes affecting/E/, the mechanismof diffusionupwards(socially)throughthe communityis quite different. Whatseemsto be involvedhereis a 'drift' phonetically to the back of the characteristic middle-class realization. The datapresented heresuggestthat socialnetworkstructure is implicated in processesof linguistic changein at leasttwo ways.First,a strongcloseknit networkmay be seen to functionas a conservative force,resisting pressures to changefrom outsidethe network.Those speakers whose ties are weakest are those who approximate least closely to vernacular norms,and are most exposedto pressures for changeoriginating from outsidethe network. Second,a detailedsociolinguistic analysisof [?] raisingand [a] backingprocesseswhichhave a commondialectalpoint of originbut have takenon very differentsocial values in their new urban context- suggeststhat the VERNACULAR speakers associatedmost stronglywith the innovationare in eachcase thosefor whomthe vowelfunctionsleastprominently as a network marker.It is as if a strongrelationship betweenthe networkstructure of a given group and choice of phonetic realization of a particularvowel disqualifiesthat group from fulfillingthe role of innovatorswith respect to that vowel. Conversely,it may be the case that dissolution of the language/networkrelationshipwith respect to a group of speakersis a necessary conditionfor that groupto fulfilthe role of linguisticinnovators. Both of these observationssuggest that since the variable NETWORK STRUCTURE is implicatedin a negative way in linguisticchange, a closer examination of WEAK networkties would be profitable. For it mightwell be



that it is speakerswho lack strongnetworkties or are loosely attachedto closeknitgroupswho are characteristically linguisticinnovators. The problemis that a generalweaknessof social networkanalysisis its superior abilityto handleCLOSEKNITties as opposedto weak,diffusetypesof networkstructure. This difficulty arisesfromthe fact that personalnetworks are in principleunbounded;the numberand strengthof ties whichbind an individualto others are not, in the last analysis, definable.However in closeknitterritorially definedgroupsit is possibleto treatpersonalnetworks AS IF they wereboundedgroups(see Milroy, I980: Ch. 3) whereas in socially andgeographically mobilesectorsof societythisis not feasible. Ourownwork has reflected this in that it has concentrated on the functionof closeknitties, observedwithina definedterritory, as an important mechanism of language MAINTENANCE. Yet, it is evident that a very large number of speakers, in cities, do not have personalsocial networksof this type. We particularly havesuggested that,in Britishsocietyat least,closeknitnetworksarelocated primarily at the highestand the loweststrata,with a majorityof sociallyand mobile speakersfalling betweenthese two points. (But see geographically Kroch, M.S., for an interestingstudy of a closeknitupper-class AMERICAN network).Significantly, Labov and Kroch have noted that in the United Stateslinguistic changeseemsalwaysto originate anddiffusefromsomepoint - neverfrom the highestor the in this centralarea of the social hierarchy
lowest social groups (Labov, I980; Kroch, I978).

Thus,despitethe difficulties of studyinglooseknitnetworktiesin the outer cityusingthemethodsadoptedin theinner-city areas,a searchfor someother meansof followingthroughtheirevidentassociationwith linguisticchange seemedwell worthwhile.5 This cannot be accomplished by analysingstatisticallyrelationships betweenlanguageand network,as was possiblein the innercity communities, simplybecausethereis no obvious way of characterising quantitatively looseknituniplexties whichextendovervast distances and are often contractedwith large numbersof others. Indeed, such an undertaking mightbe neitherpossiblenor desirable, giventhe verydifferent role fulfilledby the closeknitgroupsat eitherend of the social hierarchy in maintaining polarisedsets of linguisticnorms.It is certainlynot clear that quantitative examination of the looseknitnetworks contracted by a majority of speakers in the centreof thathierarchy wouldbe particularly illuminating.
werethe practicalones whichmight be predicted. [5] The initialdifficulties We found that network tiesof outer-city in thekeynetwork individuals sectors of kin,friendship andwork oftenstraggled overextensive areas.Conversely, tiesof neighbourhood, whichwerecrucial in the innercity, often seemednot to be significant, sincepeoplefrequently hardlyknew theirneighbours. Thus,in the innercity, whereties weredenseand territorially bounded, it seemedreasonable to studycommunity linguistic normsusinga networkmodelwhich was itselfpartof a theoryof language maintenance. But it was not at all clearwhatkind of hypothesis we mightderivefrom a comparable studyin the outer-city areas- or even whatmightconstitute a comparable study.




We thereforeproceed to examine the relationshipbetween looseknit networktiesandlinguistic changein quitea different way. First,a theoretical model of the socialfunctionof 'weak' networkties is presented; second,we of innovatorsin general,and on this basis look at the social characteristics suggesta new model of linguisticinnovationand diffusion. 5.

The discussionin this section depends heavily on a suggestivepaper by as important Granovetter (1973), who sees 'weak' ties betweenindividuals linksbetween micro-groups (small,closeknitnetworks) andthe widersociety. Perhapsit is best at this point to graspthe nettle, and attempta definition of what is meantby 'weak' and 'strong' ties, for this contrastcannoteasily be characterized Granovettersuggests the following: 'the quantitatively. strengthof a tie is a (probablylinear)combinationof the amountof time, the emotionalintensity,the intimacy(mutualconfiding)and the reciprocal
services which characterise a tie'
(I 36I).

Note that by this measure multiplex

ties - i.e. thosewithmultiple content- wouldbe countedas relatively strong; the notion of multiplexity was an importantbasis of the networkstrength measuresused in the Belfastinner-citystudies. Granovetter'sdefinitionis probably sufficientto satisfy most readers' intuitivesenseof whatmightbe meantby a 'strong' or 'weak' interpersonal tie, corresponding as it (approximately) does to an everydaydistinction betweenan 'acquaintance' and a 'friend'.It is certainlysatisfactory for our purposehere. Granovetter remarks thatmost networkmodelsdealimplicitly with small, well-defined groupsWITHIN whichmanystrongties arecontracted (cf. p. 363 above).His fundamental is thatweaktiesBETWEENgroupsprovide argument bridgesthroughwhichinformation and influence arediffused, and thatweak ties are more likely to link membersof DIFFERENTsmallgroupsthan strong ones, whichtend to be concentrated WITHIN particular groups.Thus, while strong ties give rise to local cohesion, they lead, paradoxically, to overall fragmentation. Only weak ties can form a bridge between cohesive groups, for the followingstructural reason(whichGranovetter expresses as a hypothesis and initiallysupportsby aprioristic argument ratherthan by adducingempirical evidence): If we considertwo arbitrarily selectedindividuals, A and B and the set S,
consisting of C, D, E ... of all persons who have ties with either or both of

the relationship them,the stronger betweenA and B, the morethe networks of each are likely to overlap.Extensiveoverlap,whichwill inhibitthe flow of NEW information betweenA and B, is predicted to be least whenthe A-B tie is absentand to increasein proportionto its strength.This relationship betweennetworkoverlapandstrength of tie resultslargelyfromthe tendency 364



for exampleif A spends for strongties to involvemore time commitment; a large proportionof his time with B, it is likely that this time investment will bringhim ultimatelyinto contactwith the individuals C, D and E who initiallyformedpart of B's network.Conversely,the networksof A and B arelesslikelyto overlapif the tie betweenthemis weak,and so we mayderive the more general(and for our purposemore important) principlethat links BETWEEN closeknitgroups are normallyWEAKties betweenthe individuals them.Theseweakties betweennon-overlapping who havecontracted groups provideimportantbridgesfor the diffusionof innovations. of strongand weak variouslikelyand unlikelyconfigurations Examining of eachwhich notes that individuals ties, Granovetter varyin the proportion they contract.While not all weak ties functionas bridgesbetweengroups, all bridgesmust, Granovettersuggests,be weak ties. For the sake of the argument, a bridgeis definedas the ONLYroutethroughwhichinformation flowsfromA to B, or fromanycontactof A to anycontactof B (seeFigure2).

A--------BI K F

Figure 2 Weak ties; A bridge between two networks. ----,

,strong ties.

which Granovetter's interestis in exploringthe interpersonal mechanisms connect small groupsto each other and to a largersociety, and his model willflowthroughweaktiesrather thatinnovationandinfluence predicts than of tie andnetworkoverlap betweenstrength strongones. It is the relationship whichleads him to suggestthat NO STRONG TIE CAN BE A BRIDGE. And while it must be acknowledged that in practicethereis likelyto be more than one link betweengroupsof any size, the principlethat these links are likely to be weakis of greatimportance here.Weakintergroup ties, by Granovetter's arelikelyto be criticalin transmitting argument, innovations fromone group to another,despitethe commonsense assumption that STRONG ties fulfillthis role (see for exampleDownes (I984: 155) who suggeststhat networksmay be importantin developinga theoryof linguisticdiffusion,but assumesthat it is strongties whichwill be critical). AlthoughGranovetter's principle mightat firstseemcounter-intuitive and a littlethoughtconfirms thatit worksout wellempirically. paradoxical, First of all, it is likely (in the networksof mobile individualsat least) that weak ties are more numerousthan strongties. Second,it is clearthat manymore individuals can be reachedthroughweak ties than throughstrong;consider


for examplethe numberof contactsmadeby a salesmanin the courseof his business,duringwhichhe buildsup an elaboratestructure of bridges.Consider also the elaborate bridges set up by participants at academic whichlink the cohesivegroupsassociatedwith eachinstitution. conferences, It is via these bridgesthat new ideas pass from one institutionalgroup to another.Conversely, information relayedthroughstrongties tendsnot to be innovatory;as Granovetter remarks,'If one tells a rumourto all his close friendsand they do likewise,manywill hearthe rumoura secondand third time, since those linked by strong ties tend to share friends' (1366). But it is evidentthatgenuinediffusion of the rumour will takeplaceif eachperson tellsit to acquaintances withwhomhe is onlyweaklylinked;theyin turnwill transmit it to a largenumber of non-overlapping groups,so thatthe 'retelling effect' will not occur. It has often been noted (see, for example,Turner,I967) that a closeknit networkstructure will usuallynot survivea changeof location,andit is clear in generalthat socialor geographical mobilityis conduciveto the formation of weakties. Moreover,a mobileindividual's weakties arelikelyto be much morenumerous than his strongties. If a man changeshis job, he is not only movingfromone networkof ties to another,but establishing a link between eachrelatively cohesivegroup.Thus,mobileindividuals who arerichin weak of theirmobility)relatively ties, but (as a consequence to any given marginal cohesivegroupare, it is argued,in a particularly strongposition to diffuse innovation. Note thatthiscontention is in linewiththetraditional assumption by historiansof languagethat the emergent,mobile merchantclass were largely responsiblefor the appearanceof Northern (and other) dialectal innovationsin EarlyModern(Standard) English(see, for example,Strang, 1970: 2I4 f.; Ekwall, I956; Baugh & Cable, I978: 194); if it is correct, Granovetter's that the overlapof two individuals'socialnetworks principle variesdirectlywith the strengthof theirtie to one anotherhas considerable implicationsfor any theory of diffusion. (Strengthof tie is of course a continuous variable although'weak' and 'strong'tieshavebeentreatedhere as if they werediscrete.)It mightappearthat this relatively clearhypothesis could easily be supportedor disconfirmed empirically;but unfortunately networkor sociometric studiescannot easilybe used directlyas a sourceof corroboratory (or disconfirmatory) evidencesimplybecausetheir research designusuallyentailsrelativeneglectof weak ties. Thus, for example,when personsareaskedto nameothersfromwhomtheyhavereceived information (or friendship, as in Labov's'lames' study)the numberof permitted choices is usuallyrestricted so that the namingof weak ties is effectively inhibited. Even if the research of personswith weak ties designpermitsidentification to specifiedothers, as did our own (see Milroy, I980, for details), it is extremelydifficultto study those ties just because they ARE weak and to EGO. perceivedas relatively unimportant 366



Fortunately, empirical evidence to support Granovetter'smodel has - notablyRogers'and Shoemaker's fromelsewhere emerged (I97I) studyof the diffusionof aroundfifteenhundred innovations.Somegeneralprinciples can be extractedfrom this large body of empiricalevidencewhich tend to supportGranovetter's contentionthat innovationsfirst reach a group via weak ties. A distinction in somedetailby theauthors is between discussed INNOVATORS and EARLY ADOPTERS of an innovation. This distinctionturns out to be important whetherthe innovationis agricultural (the introduction of hybrid seedcorn to an Iowancommunity); to technological (machinery engineering firms);educational(new methods of mathematics teaching);or concerned with publichealth (introducing the habit of boilingcontaminated waterto Peruvian villagers). Thereis evenan earlylinguistic studyof the introduction of lexicalinnovationsto an oilfield(Boone, I949). All of these studies,and verymanymore,confirmthe principle that INNOVATORSare marginal to the group adoptingthe innovation,often being perceivedas underconforming to the point of deviance. The EARLY ADOPTERS of the innovationare, on the other hand, central members of the group,havingstrongtieswithinit, andarehighlyconforming to group norms;they frequently providea model for other non-innovative members of the group.Afterits adoptionby thesecentralfigures(frommore marginalpersons),an innovationis typicallydisseminated from the inside outwardswith increasing speed,showingan S-curveof adopterdistribution throughtime. Whileit is clearthat linguisticinnovationsdifferin a number of respectsfrom, for example,technicalinnovations(see Trudgill,I983: 63, for a discussion), theydo not appearto be DIFFUSEDby mechanisms markedly different fromthosewhichcontrolthe diffusion of innovations generally. For linguisticinnovationsalso show this characteristic S-curveof distribution throughtime (see Chambers & Trudgill,I980: I76-I8i; Bailey, I973). Bearingin mindthe norm-enforcing character of a groupbuilt up mainly of strongties, and its consequentlack of susceptibility to outsideinfluence, we can see why innovatorsare likely to be personswho are weaklylinked to the group. Susceptibility to outside influenceis likely to be greaterin inverseproportionto strengthof tie with the groupand by implication also in inverseproportionto susceptibility to norm-enforcing pressurefrom the group. Thus, where groups are linked by many weak ties they will be to innovationpartlyfor this (social)reason,and partlybecause susceptible innovationis for structural reasonsunlikelyto be transmitted via a strong tie (see pp. 364-365 above). Personsat the centreof a norm-enforcing group (i.e. personswho share strong ties within it) will, as a corollary, not be susceptibleto outside pressures.Becauseof the investmentin time and commitmentneeded to maintainthese strong ties, they will almost certainlylack opportunities to


form many bridges(weakties) with othergroups.Thus, typically,for these variousreasons,innovators(as opposed to early adopters)will be persons with many weak ties to other groups. marginalto theircommunity, It appearsat firstto be difficult to explainhow thesemarginal innovators could diffuseinnovationssuccessfully to centralmembersof the group;but two relatedpointscan help us here.First,in view of the verygeneralfinding of sociolinguistic researchthat the prestigevalues attachedto languageare to tapdirectly, oftenquitecovertanddifficult wemaysuggestthata successful innovationneedsto be evaluated positively,eitherovertlyor covertly.Thisis of coursea necessary but not a sufficient conditionfor its ultimateadoption, and is bindingon non-linguistic innovationsalso. wemaysurmise withGranovetter thatsinceresistance to innovation Second, is likelyto be greatin a norm-conforming group,a largenumberof persons will have to be exposedto it and adopt it in the earlystagesfor it to spread successfully. Now weak ties are, in a mobile society, likely to be very much more numerous thanstrongties, and someof themarelikelyto functionas bridges to the groupfrom whichthe innovationis flowing;thus an innovationlike the Cockney merger between /v/:/6/ and /f/:/0/ reported in teenage Norwichspeakers by Trudgill(I983: 73) is likely to be transmitted through a great many weak links contracted between Londoners and Norwich speakers.Quite simply, before it stands any chance of acceptanceby the centralmembersof a group,the links throughwhichit is transmitted NEED to be numerous(cf. Granovetter, I973: 1367). Returning to our firstpoint, we assumethat some kind of prestige,either over or covert,is associatedwith the innovation.In other words,Norwich whetherthey aremarginal or centralto theirlocal groups,in some speakers, - moredesirable senseview vernacular London speechas desirable than the speechof othercities.6Again,followingthroughthe arguments in presented this section,we suggestthat personscentralto the networkwouldfinddirect innovationa risky business;but adoptingan innovationwhich is already on the edges of the group is much less risky. Thus, instead of widespread askinghow centralmembers of a groupare inducedto acceptan innovation frommarginal members, we can viewthis as a sensiblestrategyon theirpart. In orderto adopt an innovationwhichis seenas desirable, theydiminishthe risk of a potentiallydeviantactivityby adoptingit from personswho are alreadylinkedto the group,ratherthan by directimportation.
[6] The merger between dental and labio-dental fricatives has been noted in the speech of Sheffield adolescents also. By the reasoning which we are using here, we must assume first that weak ties exist between Sheffield adolescents and London speakers and second (crucially) that London speech has some kind of prestige for Sheffield speakers. Although we cannot at this stage enumerate the factors which give rise to covert attitudes of this kind, it seems reasonable to suggest that for reasons of (for example) local loyalty, combined with perceptions of relative autonomy, not all cities will share them.




We are now in a position to relatethe substantive points emergingfrom this discussion to earlier argumentsconcerningthe relationshipbetween social networkstructure and linguisticchange.It is clearthat the link noted earlierbetweenthe dissolutionof closeknitnetworksand the susceptibility of Granovetter of a groupto linguisticchangefits in with the observations and Rogers and Shoemaker.Further, we showed that the groups most of the innovative raised/?/ andbacked stronglyassociated withthe diffusion to be those verygroupswho tendedleast to use these /a/ variantsappeared phonologicalelementsas 'networkmarkers'.It is likelythat the sociallocus of the use a givengroup of the innovationsis, at leastin part,a consequence is making(or failingto make)of them as networkmarkers. of theactuation If wereturn to Labov'sdiscussion problem (seepp. 342-343 above),it is clearthat the modelelaborated heredoes not entirelyagreewith his accountof the individualswho actuatelinguisticchange,i.e. introduce an innovationto a definable as persons group.Recallthat they aredescribed who have high prestigeand a largenumberof ties BOTH INSIDE AND OUTSIDE the small local group. They do not sound at all similar to the typical innovator,described by Rogersand Shoemaker as underconforming to the point of deviance. One seriousdifficulty no easy way appearsto be that thereis apparently forempirical studiesof linguistic changein progress (particularly phonological change)to makethecrucialdistinction between INNOVATORS(whoaresocially marginal) and EARLY ADOPTERS (who occupy a central position in the network).We can only trackan innovationthroughhistorical, geographical and social space,finallylinkingit with a specificgroup.It is not clearhow, without being able to pinpoint the time of the first introductionof an innovation to a community,we could identify this group confidentlyas innovators on the one handor earlyadopters on the other.Butit is important in principleto distinguishbetweenthe two groupsand it seemslikely that phonologicalinnovationwill alreadyhave begunto diffusethroughoutthe groupif it is sufficiently well established to be observable. We shall shortly discussthis questionin relationto the groupwhichappearsto be leadingthe changeto /a/ backingin the Clonard,Belfast. Most probably,the personsdescribed by LabovareEARLY ADOPTERS. But thereis still a problemin that it is not at all clearhow theirgroup-internal ties could be strongwhen they have simultaneously a largenumberof such ties (relative-toothers)and a high proportionof ALL theirties outside the group.Onedifficulty in assessingLabov'sworkfromthe perspective we have adoptedhereis thathe seemsto relyfundamentally on theexplanatory power of the notion of the PRESTIGEof the innovators,payingless attentionto the contentor structure of INTERPERSONALLINKS. We have argued,on the other hand, that although a successfulinnovation needs in some sense to be positively evaluated, generalizations canbemadeaboutthesocialmechanisms controlling innovationanddiffusion quiteindependently of theprestige value 369


attached to any given innovation (see the discussion of /a/ and /?/ in Section 4 above). Despite these difficultiesarising partly from differencesin theoretical orientation, the persons described by Labov do in fact correspond reasonably closely to Rogers' and Shoemaker's account of highly conforming individuals with strong ties inside the group who serve as models to others. What is clear is that the marginals who are identified as typical innovators are precisely the kind of individuals to whom Labov, in the best tradition of small-group studies, is likely to pay little attention. In fact, they closely resemble the famous 'lames' of the Harlem study, who belong centrally neither to the community youth networks nor to other networks outside the community. They are marginal to both, providing a tenuous link between them. We thus emerge with a model of linguistic innovation and diffusion which at first sight seems counter-intuitive, although we have tried to suggest at various points that it agrees reasonably well with historical and sociolinguistic observations. Specifically, it is suggested that at the small group level linguistic innovations are transmitted across tenuous and marginal links. Thus, for the very reason that persons who actuate linguistic change may do so in the course of fleeting, insignificant encounters with others occupying a similarly marginal position in their social groups, direct observation of the actuation process may be difficult, if not impossible. What we most probably CAN observe is the take-up of the innovation by the more socially salient EARLY

At the macro-level, societies undergoing social processes which entail social and geographical mobility and the dissolution of closeknit networks (processes associated with industrialization) provide the conditions under which innovations can be rapidly transmitted along considerable social and geographical distances (see Trudgill (1983, Chapter 3) for a relevant study of geographical diffusion). Bearing in mind the difficulty of studying directly the early stages of an innovation, we proceed now to assess the usefulness of the model developed here. Specificproblems associated with innovation and diffusion are discussed, first at the level of small groups and then at the level of larger national communities. 6.

The possible explanatory value of a theory of weak ties can be considered in relation to observed patterns of language variation. In certain cases, these patterns are difficult to explain in terms of the usual assumptions about linguistic diffusion, viz. that it is encouraged by frequency of contact and relatively open channels of communication, and discouraged by boundaries of one sort or another, or weaknesses in lines of communication (see, for example, Labov (I974b) for an empirical study which links the location of dialect boundaries with a trough in north-south links). 370



In Belfast,two instancesstand out which are difficultto explainin this of the way. They are (i) the social configuration common-sense apparently east of the city into the Clonard, spreadof /a/ backingfrom the protestant and (ii) the city-wideyoungergeneration a West Belfastcatholiccommunity consensuson evaluationof variantsof the (pull)variable(as againstgreater to as Details of these variables,referred in the older generation). variability in Milroy& Milroy(1978),and details (a) and (A),aremost easilyaccessible for /a/ are also givenin Table4 (above,p. 357) and Figure3.
(a) index score 350 300 250


-H H


1 Men 40-55

l Women 40-55

Men 18-25

Women 18-25

Figure 3 Backing of /a/ in Ballymacarrett, the Clonard and the Hammer.

Thebackingof /a/, as we haveseen,is led by EastBelfastmales:thismuch of the detailsin Table is indicatedby Table4. However,as the significance in here the lightof the general we discussthembriefly to interpret, 4 is difficult of the representation argumentof this paper. Figure 3 is a diagrammatic it shows the communities; spontaneousstyle patternfor all threeinner-city change in progress(Labov, 'cross-over'patternthat tends to characterize 1972a). The change appearsto be carried,not by West Belfastprotestant males (as might be expected), but by the younger FEMALE group in the This is the groupthat exhibitsthe cross-over CATHOLICClonard community. pattern. high incidenceof It may be objected,however,that thereis a moderately backingamongst older Clonardmales, even though this group shows no (on which see below). But it is the young Clonard stylisticdifferentiation THEGENERALLY EXPECTEDPATTERNS.Amongstthem,the femaleswho REVERSE the incidence city-widefemalepattern(awayfrom /a/ backing)is reversed: femalegroups, in thisgroupis higherthanin olderandyounger of /a/ backing - higher higherthan in the olderClonardfemalegroup,and - surprisingly in the Clonardarea. When than amongsttheir youngermale counterparts measuredagainstother groups,these young womenappearto be reversing a trend. is additionally takeninto account,it is clearthat WhenSTYLISTICpatterning


this young female group is the only Clonard group with significant stylistic differentiation on the East Belfast model (see Table 4). Their usage is innovatory in West Belfast in that the social value attached by them to the variants is the same as the social evaluation evident in the East Belfast data, but not well established in the west of the city. Thus, while superficial consideration of the figurescited might suggest that the young Clonard female pattern is modelled on older Clonard males, such an explanation would not account for reduction of /a/ backing in other groups, nor would it account for the use of (a) as a stylistic marker by the Clonard girls. The social barriersthat inhibit contacts between working-class communities have been well described (see for example a discussion of this work in L. Milroy, ig80) and it is clear, as Boal (1978) has shown, that the intercommunity conflict in Belfast has strengthened these barriers. In fact, the major traditional sectarian boundary in West Belfast is now marked physically by a brick and barbed wire structure, which is described by the military authorities, apparently without intentional irony, as 'The Peace Line.' The puzzle is, that an East Belfast pattern can be carried across these boundaries, evidently by a group of young women whose physical movements and face-to-face contacts have been constrained from a very early age. It is clear that the diffusion of patterns of /a/ backing from east to west, progressing in a linguistically and stylistically ordered manner, is a continuation of the long term shift in the Belfast vowel system (together with the social values attached to it) described in Section 3. That this shift is continuing apparently unhindered across the iron barriers, both physical and psychological, which separate protestant East and catholic West Belfast, is a fact for which up until this point we have not felt able to propose any principled explanation. The continuation of the change may now be considered in terms of the claim that INNOVATORS who are marginal to a group introduce innovations, to EARLY ADOPTERS who are central figures within that group. The innovation is likely to be transmitted by means of weak, rather than strong, ties. In addition to scoring high on /a/ backing - a score reflecting both quantitatively and qualitatively the speaker's choice of realization - the Clonard girls scored extremely high on the Network Strength Scale, which was designed to measure relative centrality of position in the closeknit group. Hence, they resemble Rogers' and Shoemaker's EARLY ADOPTERS rather than, strictly, INNOVATORS. As described in L. Milroy (I980), the girls were all in employment and were all associated with the same rather poor city-centre store. This store was located in North Street, a shopping area on the sectarian interface which served both protestants and catholics, mainly those living west of the river (the girls' male counterparts contrasted sharply with them in being unemployed and scoring low on network strength). We need to emphasize at this point that when we argue, with Boal, that 372



there are few ties betweenworking-class groups in Belfastwe mean more or properlythat thereare few STRONG TIES such as those of kin, friendship are work, particularly acrossthe sectariandivide.But there plenty of weak ties(to whichwe havehithertopaidlittleattention) between,on the one hand, West Belfastcatholicsand protestantsand on the other, betweenEast and meet West Belfastprotestants.Some of the settingsin whichthey regularly are(as reported in the innercity study)shops,hospitalwaiting by informants rooms, social securityofficesandjob centres.The Clonardgirlsworkingin the shop would be extremelywell placed to adopt innovationstransmitted weaklinkswith by personson theedgeof theirnetworkwho in turnprovided othercommunities. We havealreadyarguedthat if an innovationis to stand anychanceof adoption,theseweaktieencounters wouldneedto be frequent that is, they wouldneed to be with a largenumberof back [a]users.It may be surmised, giventhe numberof serviceencounters in the shop in any one encounters withback[a]userswho transmit theinnovation day,thatweak-tie will greatlyexceedin numberstrong-tieencounters with non-back[a]users. Hence the capacity of innovation-bearing weak ties to compete with innovation-resisting strongties. If we havea theoretical perspective suchas the one developedhere,which explicitlypredictsthat an innovationwill be transmitted throughweak ties, perhapsin casual serviceencountersperceivedby participants to be of no affectivevalue, the back [a]diffusionproblemdissolves.The problemarises in the first place only if we assume that strong ties must be involved in of innovations; forin thatcase,a searchfor an explanation diffusion in casual in waitingrooms,shopsanddole queueslooks likethe worstkind encounters
of ad hoc-ery.

The secondpuzzleconcernsthe (pull)variable,whichis associatedwith a small number of lexical items alternatingbetween the two phonological classes/u/ and /A/ - examplesarepull, push, took, shook,foot. Thecomplex historyof this subset(see J. Milroy I980 for details)is apparently reflected in greatinstabilityamongall but the youngerinnercity speakersboth with regard to the specific lexical items assigned to one or another of the phonologicalsets, and with regardto the social value assignedto the [A] variant. Thus,forexample, somespeakers explicitly stigmatized [A] realizations of itemslikepullandpush,whileothers,in so faras theyusedan [A] realization for readingthem on a word list, apparently consideredformslike [pAl] and in the wereparticularly [pAJ] to be 'correct'.Overall, [A] realizations frequent Clonard,especiallyamongthe olderwomenand even in carefulstyles. Whenwe turnto the youngergeneration, the picturechangesradically, as can be seen from Figure4. What this diagramreflectsis a processof lexical between[A] and [u]realizations are diffusion,wherebyitemswhichalternate graduallystabilisingin the /u/ set. But the few items which continue to alternate have assumedveryconsiderable the [A] sociolinguistic significance, realizationsbeing perceivedas stronglysymbolicof Belfastworkingclass 373


(A) %



Men 40-55

Women 40-55

Men 18--25

Women 18-25

Figure 4 Distribution of the (pull) variable (% of [A] variants are shown) by age, sex and area in inner city Belfast.

usedby youngmen very languageandculture.As such,they areconsistently muchmore than by young women,as Figure4 shows. The puzzleis of coursehow youngpeoplelivingin the closedcommunities of Ballymacarrett, Clonardand Hammer,whosecontactwith othersoutside their areas has been only of a very tenuous kind, have come to reach consensuson the social value to be assignedto the two cross-community variants of the (pull) variable. Paradoxicallytheir parents, who formed friendships much more freely across sectarianand communityboundaries (untilthe beginningof the civil disorders in I969), still showedconsiderable variationboth in the use and the evaluationof the variable.This variability was apparentlyin the first place partly a consequenceof the different phonologies of various hinterlanddialects. Yet, the process of lexical diffusionand the absorptionof the (pull)variableinto the regularsociolinguistic structure of Belfast's urban dialect continued unhindered, unaffected apparently by the inabilityof the youngergeneration to contract any strong interpersonalties across the sectariandivide. For it is these youngsters,and not their parents,who show dramaticagreementon the form whichthese sociolinguistic patternsshouldtake. Althoughthereis stilla greatdealto explainaboutthechanging distribution of a complex phono-lexicalset like (pull), the question of how city-wide consensus on its use andevaluation wasreached by theyoungerspeakers does not now seempuzzling.Like the diffusionpatternof /a/ backing,the (pull) problemdissolvesif we accept that weak ties are the normalchannelfor diffusionof innovations. 374



Having discussedthese details of change and diffusion in present-day Belfast,we turn in the next section to the place of weak ties in long-term languagechange.

It is well knownthat in the courseof historysome languageshave changed moreradicallythan others.In the Indo-European family,certainlanguages, such as Lithuanian,are acknowledged to be highly conservative,whereas others(e.g. English, havediverged Dutch,French,Portuguese) verymarkedly from theirancestral in the historyof certainlanguages forms.Furthermore, there have been periods of rapid change and periods of slow change. A comparison of the socialandcultural conditionsobtainingin periodsof slow and rapidchangeshouldcast light on the social motivationof changes. have been adducedto accountfor large scale linguistic Many arguments changes;for example,substratum theoriesand accountsof lexical,syntactic and phonological Cultural borrowing. factorshavealso beendiscussed, such as languagecontactfollowingconquestand settlementof alien speakers.In recentdecadesmuchattentionhasbeenpaidto pidginization andcreolization (Todd, I974), and pidginlanguagesare of course the paradigmatic case of linguisticinstability;they can changevery rapidly.As arguments based on substratum, conquest,etc., are not uniformlyapplicableto all situations,it maybe thata moregeneral condition(in linewiththe argument of thispaper) can be proposed,that will encompassthese variedsituations.This can be statedas follows:

We can seek supportfor this hypothesisby comparing two languages that have changedat very different rates. Amongstthe Germanic languages,Icelandicand Englishprovidea sharp contrast in rate of change and degree of variation.WhereasEnglish has changedradically sincethe twelfthcenturyand has at all recoverable periods exhibited gross dialectal variation, Icelandic has altered little since the thirteenth century and reportedly shows very little dialectal variation. Icelandic maintains a fullinflexional systemfor case,number, gender, person, tenseand mood; phonological changehas beenslight,involvingtwo mergers of low functional yieldand veryminorconsonantchanges;phoneticchanges include diphthongization of long vowels and some allophonicchanges in consonants and vowels, but it is not clear how far these had already progressedin the Middle Ages (some fifteenth-century spellings already indicatediphthongization of certainlong vowelspresumably sometimeafter 375


the changes had occurred). Change in English, on the other hand, is quite radical - amounting to a typological change from a highly inflected to a weakly inflected language. There are also many phonological changes, word-order changes and partial relexification from Romance and Classical sources (for a brief history see J. Milroy, I984a). Notice that the geographical isolation of Icelandic (although relevant) cannot be a sufficient explanation for its long-term conservatism. If geographical isolation were the most important factor, we should expect the dialects of Icelandic to have diverged considerably. Iceland is comparable in size with Britain, but the centre of the country is glaciated, and settlements are scattered around the coastal areas. The climate and terrain are such that in the Middle Ages little communication was possible in the winter months (conventionally October to April). According to the Icelandic sagas, the journey to the main assembly at Thingvellir could take weeks. In Hrafnkels Saga, the hero's journey from eastern Iceland to the west is described, and the writer comments: Sui6r6r Fljotsdal eru sjautjaindaglei6ir a'Dingvdll (South from Fljotsdal it is a seventeen day journey to Thingvellir). Hrafnkel's rival, Saimr,had an even longer journey: Ok f6rsk honum jvi seinna, at hann aittilengra lei6 (And his journey was so much slower in that he had a longer route). A theory of change based mainly on the separation of communities would surely predict that varieties would diverge rapidly in these conditions. Our hypothesis on the other hand predicts that if widely separated communities maintain the same linguistic forms, ties between them must in some sense be strong, and evidence from the Icelandic family sagas (c. 1200-1300) seems to bear this out. Iceland was colonized in the late ninth and tenth centuries by independently minded Norwegians, some of whom had settled in the Orkneys, Shetlands and Hebrides prior to their emigration to Iceland. There was little social stratification in the Icelandic Commonwealth: there was no aristocracy, and the feudal system had no effect until after the annexation of Iceland by the Norwegian crown in the late thirteenth century. Although Christianity was accepted officially in IOOO,the temporal power of the Church appears to have been less than elsewhere. In Icelandic writings, the early missionaries are represented as thugs, and the status of priests seems for some time to have been hardly better than that of farm-servants. In short, institutional power seems in general to have been weak enough to allow informal kinds of social organization to flourish. The thin population was widely distributed, but an early form of quasidemocratic government evolved. The country was divided into districts, and, in these, assemblies were held at which attempts were made to settle disputes 376



andpronounce judgmenton wrong-doers. Everyyear,the nationalassembly washeldat Thingvellir, nearReykjavik, (theAlthing) andpeoplewouldtravel verylongdistances to this.Accountsin the sagassuggestthatthisinstitutional was not very successfulin settlinglegal disputes,and it was superstructure certainly unableto carryout punishments. In orderto get redress for offences, peoplewerein practice on the supportof theirfamiliesand whollydependent friendsand those who had obligationsto them.It was veryimportant, in the absenceof stronginstitutional power,that strongties shouldbe maintained with those who mighthelp in a time of need. The assemblies were,in practice, a meansof maintaining strongties across long distances,and the sagas furthershow the great importancethat was attachedto personalidentity,kin and friendship. When a new character is introducedby the saga-writer, a paragraph or more is typicallydevotedto naminghis parentsand grandparents distantancestors), (and sometimes his brothersand sisters,his wife and family (and sometimesother relatives). Whena stranger appearsin the story,he is oftenquestioned abouthis name, his home, his relativesand his status.WhenSaimr, in Hrafnkels Saga, meets a stranger (whomaybe ableto assistin a law-suit), he askshis name,whether or not he is a local leader(godordsmadr) or farmer(bondi), who his brothers are,and so on. The stranger's repliesgivemoreinformation thanthe modern readermightthinknecessary. In Hrafnkels Saga, the stranger tells Saimr that his brother's nameis Pormo6, that Pormo'6r livesat Gar6aron Alptanesand thatheis married to Dordis, whois thedaughter ofD0rolfr,sonof Skalla-Grim, from Borg.This kind of exchangeof information is typicalof the saga; it is also typicalof communities thatdependon maintaining strongnetwork links. Similarexchanges,the purpose of which is to declare identity, political affiliationand personalrelationships, were reportedby informantsin the Belfastproject (L. Milroy,I980: 55); theseinformants wereattached to strong territorially based social networks. The conservatism of Icelandicand the relativelack of variationin that language maytherefore be attributed largelyto the greatpractical importance attached to maintaining strongly established kinandfriendship networks over long distances and through many generations. As in the low-status
communities described by Lomnitz (I977) and discussed by L. Milroy (I980:

70 ff), the patternsof exchangeand obligationimposed by such network structures ensurespracticalsupportin time of need. Such a social structure (basedon informallinks)could flourishin medievalIcelandbecauseof the inabilityof pan-European institutions (the Churchand the feudalsystem)to establishtheirpowerfully.Oneof the resultsof this informalsocialstructure is the impositionof linguisticnormson its members (in commonwith other norms). Hence the failure of the language to exhibit much change or variation,despitethe difficulties of distanceand terrain. The history of English, which is dramaticallydifferentfrom that of canhardly Icelandic, be unaffected by population history.Inearlytimes,there 377


is a history of repeated incursions. Danish settlers in Eastern England in the ninth and tenth centuries found Old English (Anglo-Saxon) well established, whereas the contemporary Norwegian settlersin Iceland found an uninhabited country. The numerous Scandinavianplace-names of Yorkshire, Lincolnshire and elsewhere strongly suggest that Danish-speaking communities survived in these areas for some time. If so, the ties contracted between the Danes and the English could not, for social and linguistic reasons, have been strong in the first place. The communication that must have taken place in the course of trade and farming seems to have been carried on in an Anglo-Danish contact language. This is indicated by the nature of the language that emerges in the Middle English texts of these eastern areas (e.g. The Peterborough Chronicle, I 137) which is an Anglo-Saxon-based language with gross inflexional loss, absence of grammatical gender, and partial reflexificationfrom Danish and Norman French. On the other hand, the English of the West Midlands around 1200 - an area largely unaffected by the Danes - provides a startling contrast. The Ancrene Wisse, for example, is morphologically conservative (in that gender and case inflexions are largely retained), and Danish loanwords are very rare. Thus, we appear to have relatively rapid change in areas where pre-existing strong networks are disrupted and where influence through weak ties is made possible: on the other hand we have a conservative language in areas of the West Midlands where Anglo-Saxon institutions remained more stable, and where neither Danish nor Norman influence was initially strong. The success of the Norman Conquest imposed a tight and organised administration on much of the country; rule was more centralized, and class divisions more fully institutionalized by the feudal aristocracy. While Iceland remained a yeoman democracy, England acquired an institutional system of social stratification. One of the effects of stratification is the creation of social distance between sectors of the population. Two developments in English may be a general consequence of social distance and weak ties. The first is the character of relexificationfrom Norman French. There is a rapid development of English/ French synonyms of the type child/infant, love/charity, board/table, stool/chair: the French synonyms tend at first to be limited to more formal social contexts. The second development is the use of the polite pronoun of address, which was marked for status and social distance: it was used asymmetrically and non-reciprocally by inferiors to superiors. Brown and Gilman (1972) estimate that the non-reciprocal polite plural pronoun entered

most European vernacularsbetween I I00 and


with French very

advanced in this respect. In Icelandic, this development is relatively late. It is unknown in the sagas of the Icelanders (c. I200-I300) for use between Icelanders, although Icelandic adventurers (around I300) are occasionally represented as addressing some European monarch with the polite pronoun. 378



It is not untilchivalric in thefourteenth sagason non-Icelandic themesappear centurythat the non-reciprocal politepronounbecomesreasonably common in literature (oftentranslated). As far as we are aware,thereis no indication that, at this date, nativeIcelanders used it amongstthemselves. A finaldevelopment in Englandthat tendedto encourage the break-up of strongties and the development of weak ties was the rise in the importance and populationof London.Londonbecamethe seat of the Court,the main commercialcity and the centre of the wealthiest part of the country. Immigration to London(Ekwall,1956;Strang,I970: 2I4 f.) was frommany areas,but largelyfrom the East Midlands(resulting in a gradualchangein the dialectfrom southernto East Midlands).The rapidinflexional loss that diffusedthroughoutthe ME periodcan be seen, not only as a resultof the influence of weaklyinflected (E. Midland)dialectson stronglyinflected ones, but as a productin Londonof the contactsituationitself,in which'mergers expandat the expenseof distinctions'(Herzog,quotedin Labov, I972: 300). In thirteenth- and fourteenth-century Iceland, there were no such developments. in searchof theirfortunehadno largetownto settle Icelanders in; they tendedto go abroadfor a time and then returnto theirruralhomes in Iceland.In such conditions,strongnetworksremainedto a largeextent intact. Thus,the contrastbetweenEnglishand Icelandicseemsto be an exemplar of the contrastbetweensocialconditionsthatencourage weakties and those that encouragestrongties. Rapidchangesin Englishseemto havedepended ontheexistence ofindividuals andgroups whoweresociallyandgeographically mobile and whose strongnetworkties wereweakenedor brokenup by this mobility.A high degreeof social distanceseemsto have resulted.Icelandic society, on the other hand, dependedin earlier centurieson the strong networkstypicalof rurallife. Hence, despitethe difficulties of climateand socialnetworks terrain, provedto be a cohesiveforce,not onlyin maintaining social norms,but also in maintaining the normsof language. We have discussedthe case of Icelandicand Englishin orderto support the generalization stated on p. 375 above that 'linguisticchangeis slow to the extent that the relevantpopulationsare well established and bound by strongties,whereas it is rapidto theextentthatweaktiesexistin populations'. Cases of conquestand colonizationare taken as relevanttypes of weak-tie situations for the reasonthatrapidchangeis oftenassociated withsuchcases. Nigel Vincent(personalcommunication) drawsour attentionto a possible Romanceanalogue.Sardinian is generally regarded as the most conservative of the Romancelanguageson a numberof counts, and this state of affairs canplausibly be correlated withthefactthataftertheperiodof Romanization (3rdcenturyB.C.), such incursionsand occupationsas therewerehad only a marginal effecton the social organization of the inhabitants of the island, and even then only in peripheral areas(see Blasco Ferrer,I984). Sicily,by 379


contrast, has a long history of conquest and colonization by Greeks, Normans, Lombards and Arabs, and linguistically Sicilian displays a good deal of innovation and stratification (see Varvaro, I98I). However, our generalization is intended to encompass any situation where contacts between people lead to the establishment of many weak ties. It therefore includes, in principle, situations where warlike incursions are not of major importance. Two further broad and overlapping types of contact spring to mind. One is the peaceful in-migration of populations who speak other languages or dialects. This might help to account for change in the maritime colonial languages of Western Europe (as against those of the interior), as metropolitan centres are characterized by some ethnic and social diversity; this is also clearly relevant to rapid change in large post-Industrial Revolution cities, in which much of the in-migration from the hinterland is by speakers of different dialects of the same language. The second type is sustained commercial and cultural contact (which leads in extreme cases not only to simplification but to 'language death'). These types of contact would seem to be relevant to cases like Danish (an old established maritime and colonial language closely related to Icelandic, but which has undergone rapid change); the history of that country has been characterized by quarrels with neighbouring states, but also by sustained commercial and cultural contact with these states. Therefore, we do not consider incursion and conquest to be a necessary condition in itself for rapid linguistic change. The correct generalization must account more broadly for the spread of speaker innovations through weak ties.

We have here presented a model designed to explain why linguistic change seems commonly to take place in some social conditions but not in others. Specifically, we have tried to show as explicitly as possible that innovations are normally transmitted from one group to another by persons who have weak ties with both groups. Further, at the macro-level, it is suggested that in situations of mobility or social instability, where the proportion of weak links in a community is consequently high, linguistic change is likely to be rapid. Social groups who characteristicallycontract many weak ties - and in Western society these could consist of persons who belong neither to the highest nor to the lowest social groups - are likely to be closely implicated in the large scale diffusion of linguistic innovations. These claims are supported by empirical observations. For example, it has been noted that innovations seem to hop from one centre of population to another, along main lines of communication such as roads and railways (Trudgill, I983: Chapter 3). This is to be expected if we assume first that they are carried by persons from community A who have weak ties with those in community B, and second that ties contracted in these contexts are likely to 380



be numerous.Our arguments here also fit in with Labov'sfindingthat the locus of changeis alwaysat somecentralpoint in the socialhierarchy where, we have suggested, ties are weak.For this reason,an accountbasedon weak ties seems to be at a higherlevel of generalitythan one based on class or status. Nor does the evaluative notion of PRESTIGE(overtor covert)havea central partto play in the modelpresented here.Labovis correctin his observation thatlinguistic innovations maydiffuse bothupwards anddownwards through the social hierarchy; some appearto originatewith high and some with low status groups.Conversely,a comparisonof the diffusionmechanisms and distributional patterns of /a/ and/E/ in Ulstershowsthatelements originating from the same (rural)dialect can take on, apparentlyarbitrarily, entirely differentsocial valuesin theirnew urbancontext. We assume(althoughanydiscussion is beyondthe scopeof thispaper)that perceptual and acousticfactors,as well as a rangeof moregenerallinguistic will sharplylimitthe classof possibleinnovations(cf. Weinreich constraints, et al., I968: IOO; Labov, I982: 27; Comrie,I98I: 195). But withinthe limits set by this largerclass of constraints, the notion of prestigedoes seemto be importantin explainingwhy one particularlinguisticelementis a realistic candidatefor innovationwhile others are not. We have suggestedthat the workingclassof East Belfast,who formeda kindof labouraristocracy, were particularly stronglyassociatedwith the UlsterScots hinterland fromwhich thecontemporary urban/a/ and/?/ changes haveoriginated. Thisassociation helpsexplainwhy thesephonologicalelementsand not, for example,others associatedwith mid-Ulsterdialects,have been successfully introducedand diffused.But sincethis topic also lies beyondour scopehere,we simplynote, withTarde,thatwe needto learnwhy,if a hundred innovations areconceived simultaneously, ten will spreadwhile ninetywill be forgotten(I903: 140). It has also been necessaryto distinguishsharplybetween INNOVATION (whichis the act of a speakeror speakers)and CHANGE, whichis the reflex of a successful innovation in the language system.Present day sociolinguistics (althoughsensitiveto social phenomena)is in fact stronglyorientedto a 'system' approachand has often not made a sufficiently sharpdistinction betweenthe linguisticbehaviour of speakers and the effectof that behaviour on the languagesystem. Finally, by making a furtherdistinctionbetween the INNOVATORS of a linguisticchangeand the EARLY ADOPTERS, we have suggesteda principled reason for the difficultyexperiencedin observingthe introductionof an innovationinto a community.This may be seen as the earlieststage of a linguisticchange- at least from the point of view of the communitywhich is adoptingit. Weinreich et al. have described this actuationof a changeas 'the very heart of the matter'. However, since innovators tend to be marginal individuals at theedgesof networks whodiffuse innovationvia weak ties with others, the personswhom investigators actuallyidentifyas being


the moresociallycentral stronglyassociated witha changearemostprobably early adopters. It is important to note the limitationsof the claimspresented here.We are attempting to shed light on the actuationof a linguisticchange,noting that thereis not necessarily a one-to-onerelationship betweeneven a successful speakerinnovationand the changein the languagesystemwhichreflectsit. A singleinnovationmay triggeroff a seriesof changesin a chainshiftwhich can thenbestbe explained theinternal by examining organisational principles underlyingthe language system. It is assumed that an appropriately explanatoryaccountof language changemustsupplement the modelpresented herein at least two importantways. First,it must specifythe psycholinguistic and linguisticconstraints which limit the class of candidates for innovation.Second,it must accountfor the regularand orderlymannerin which successfulinnovationsare diffused throughoutthe system,so that eventuallythey are perceived as instancesof linguisticchange.
Bailey, C.-J. (1973). Variation and linguistic theory. Washington D.C.: Center for Applied

Baugh, A. C. & Cable, T. (1978). History of the English language, 3rd ed. London: Routledge

& KeganPaul. dellaSardegna. Max Niemeyer BlascoFerrer,E. (I984). Storialinguistica Tiibingen: Verlag. theperspective on theShankill-Falls fromI976. Boal,F. W. (I978). Territoriality divide,Belfast: In Lanegran, D. A. & Palm,R. (eds),An invitation to geography. 58-77. Boone, L. (1949). Patterns of innovationin the languageof the oil-field.American Speech 24.

in Giglioli,P. P. Brown,R. & Gilman,A. (1972). Pronouns of powerand solidarity. Reprinted Language andsocialcontext.Harmondsworth: Penguin.252-82. Bynon,T. (I977). Historical linguistics. Cambridge: Cambridge Press. University J. K. &Trudgill, Chambers, P. J. (I980). Dialectology. Cambridge: Press. Cambridge University N. (1975). Language and responsibility.Sussex:Harvester Press. Chomsky, N. & Halle,M. (I968). Thesound Chomsky, New York:Harperand Row. patternof English.
Comrie, B. (I98I). Language universalsand linguistic typology. Oxford: Blackwell.

Downes,W. (I984). Language and society. London:Fontana. Eckert,P. (I980). The structure of a long-term phonological process:the back chain shift in SouletanGascon.In Labov,W. (ed.).Locating language in timeandspace.179-2 I9. Ekwall, E. (I956). Studies on thepopulation ofmediaeval London. Lund:LundStudies in English. Gal, S. (1979). Language shift. New York:Academic Press. Granovetter, M. (I973). Thestrength of weakties.AmericanJournalof Sociology 78. I360-I380. Gregg, R. J. (I964). Scotch-Irishurban speech in Ulster. In Ulster dialects -a symposium. Holywood,Co. Down: UlsterFolk Museum.I63-I9I. Gregg,R. J. (1972). TheScotch-Irish dialectboundaries of Ulster.In Wakelin, M. (ed.),Patterns
in thefolk speech of the British Isles. I09-139.

Gumperz, J. J. (I982). Discourse strategies. Cambridge: Cambridge University Press. Halsey,A. H. (1978). Change in British society. Oxford:OxfordUniversity Press. Harris, J. (I983). Linguistic change in a nonstandard dialect.Unpublished PhDthesis,University of Edinburgh.
Heilmann, L. (ed.) (1974). Proceedings of the Eleventh International Congress of Linguists.

Bologna:II Mulino. Kroch,A. S. (I978). Towards a theoryof socialdialectvariation. Language in Society 7. I7-36. 382



Kroch, A. S. (I979). Dialect and style in the speech of upper class Philadelphia. Mimeo: University of Pennsylvania. Labov, W. (1972). Sociolinguisticpatterns. Philadelphia: Philadelphia University Press. Labov, W. (1974a). On the use of the present to explain the past. In Heilmann, L. (ed.), Proceedings of the Eleventh International Congress of Linguists. Ii ii-I 135. Labov, W. (1974b). Linguistic change as a form of communication. In Silverstein, A. (ed.), Human communication: theoretical explorations, Hillsdale, New Jersey: Lawrence Erlbaum Associates. Labov, W. (I980). Locating language in time and space. New York: Academic Press. Labov, W. (I98I). Resolving the neogrammarian controversy. Lg 57. 267-308. Labov, W. (I982). Building on empirical foundations. In Lehmann, W. P & Malkiel, Y. (eds), Perspectives on historical linguistics. Amsterdam and Philadelphia: John Benjamins. 17-92. Labov, W., Yaeger, M. & Steiner, R. (1972). A quantitativestudy of sound change in progress. Report on NSF Project No. 65-3287. Lanegran, D. A. & Palm, R. (eds) (1978). An invitation to geography (2nd ed.). New York: McGraw Hill. Lass, R. (I976). English phonology and phonological theory. Cambridge: Cambridge University Press. Lass, R. (I980). On explaining language change. Cambridge: Cambridge University Press. Lehmann, W. P. (I962). Historical linguistics: an introduction. New York: Holt, Rinehart & Winston. Lomnitz, L. (1977). Networks and marginality. New York: Academic Press. Milroy, J. (1976). Length and height variations in the vowels of Belfast vernacular. Belfast WorkingPapers in Language and Linguistics I. 3. Milroy, J. (I980). Lexical alternations and the history of English. In Traugott, E. (ed.), Papers from the International Conferenceon Historical Linguistics. Amsterdam: Benjamins. 355-62. Milroy, J. (I98I). Regional accents of English: Belfast. Belfast: Blackstaff. Milroy, J. (I982 a). Probing under the tip of the iceberg: phonological 'normalization' and the shape of speech communities. In Romaine, S. (ed.), Sociolinguistic variation in speech communities. London: Edward Arnold, I982. 35-47. Milroy, J. (I982b). Some connections between Galloway and Ulster speech. Scottish Language I (Winter issue). Milroy, J. (I984a). The history of English in the British Isles. In Trudgill, P. (ed.), Language in the British Isles. Cambridge: Cambridge University Press. 5-3I. Milroy, J. (I984b). Present day evidence for historical changes. In Blake, N. F. & Jones, C. (eds), Progress in English historical linguistics. Sheffield: University of Sheffield. Milroy, J. & Harris, J. (I980). When is a merger not a merger?: the MEAT/MATE problem in a present-day English vernacular. English World-WideI. 199-210. Milroy, J. & Milroy, L. (1978). Belfast: change and variation in an urban vernacular. In Trudgill, P. (ed.), Sociolinguisticpatterns in British English. Milroy, J., Milroy, L., Harris, J., Gunn, B., Pitts, A. & Policansky, L. (I983). Sociolinguistic variationand linguisticchangein Belfast. Report to the Social Sciences Research Council, Grant No. HR5777. Milroy, L. (I980). Language and social networks. Oxford: Blackwell. Patterson, D. (i 86o). Provincialismsof Belfast. Belfast. Pitts, A. (I982). Urban influence in Northern Irish English: a comparison of variation in two communities. Unpublished PhD dissertation, University of Michigan. Rogers, E. M. & Shoemaker, F. F. (1971). Communicationof innovations(2nd ed.). New York: Free Press. Romaine, S. (I98 I). The status of variable rules in sociolinguistic theory. JL I7. 93- I19. Staples, J. H. (I898). Notes on Ulster English Dialect. Transactionsof the Philological Society.

Strang, B. M. H. (1970). A History of Engish. London: Methuen. Tarde, G. (1903). The laws of imitation (translated by Elsie Clews Parsons). New York: Holt. Todd, L. (I974). Pidgins and creoles. London: Routledge & Kegan Paul. Trench, R. C. (I888). On the study of words, 20th ed. London: Kegan Paul, Trench & Co. Trudgill, P. (ed.) (1978). Sociolinguisticpatterns in British English. London: Arnold. Trudgill, P. (I983). On dialect. Oxford: Blackwell.


JAMES MILROY AND LESLEY MILROY Varvaro, A. (I98I). Lingua e storia in Sicilia. Palermo: Sellerio. Vennemann, T. (I983). Causality in language change. Folia Linguistica Historica IV/I. 5-26. Wakelin, M. (ed.) (I972). Patterns in thefolk speech of the British Isles. London: Athlone. Williams, R. A. (1903). Remarks on Northern Irish pronunciation of English. ModernLanguage Quarterly6. 129-I35. Weinreich, U., Labov, W. & Herzog, M. (I968). Empirical foundations for a theory of language change. In Lehmann, W. & Malkiel, Y. (eds), Directions for historical linguistics. Austin: University of Texas Press. 95-195.