The mediated innovation model: a framework for researching media influence in language change

Linguistic innovations that arise contemporaneously in highly distant locations, such as quotative be like, have been termed ‘global linguistic variants’. This is not necessarily to suggest fully global usage, but to invoke more general themes of globalisation vis-a-vis space and time. This research area has grown steadily in the last twenty years, and by asserting a role for mass media, researchers have departed intrepidly from sociolinguistic convention. Yet they have largely relied on quite conventional sociolinguistic methodologies, only inferring media influence post hoc. This methodological conservatism has been overcome recently, but uncertainty remains about the overall shape of the new epistemological landscape. In this paper, I review existing research on global variants, and propose an epistemological model for researching media influence in language change: the mediated innovation model. I also analyse the way arguments are constructed in existing research, including the use of rhetorical devices to plug empirical gaps – a worthy sociolinguistic topic in its own right. 
 
 
 
Les innovations linguistiques qui emergent simultanement dans differentes regions du monde, telles que les verbes introducteurs be like par exemple, ont ete recemment appelees des ‘variantes linguistiques universelles’. Ceci ne suggere pas necessairement l'existence d'un usage mondial a part entiere, mais plutot l'invocation de themes plus larges sur la mondialisation en matiere d'espace et de temps. Ce domaine de recherche a connu un developpement regulier au cours de ces vingt dernieres annees, et en revendiquant le role que peuvent jouer les medias de masse, les sociolinguistes se sont eloignes avec une certaine intrepidite des conventions caracteristiques de leur domaine. Pourtant, ces derniers ont bien employe des methodes d'analyse sociolinguistique conventionnelles, mais inferant l'influence des medias uniquement apres coup. Force est de constater que ce conservatisme methodologique a recemment ete depasse, mais de nombreuses incertitudes subsistent quant a la structure globale de ce nouveau paysage epistemologique. L'objectif de cet article est d'examiner les travaux de recherche qui traitent de ces variantes universelles et de proposer un modele epistemologique destine a la recherche sur l'influence des medias dans les processus linguistiques du changement appele en anglais ‘the mediated innovation model’, c'est-a-dire un modele d'innovations vehiculees par les medias. Dans le meme temps, le present article vise a etudier la maniere dont les arguments sont construits dans ces travaux de recherche, par exemple par le biais de procedes rhetoriques, afin de corriger les insuffisances empiriques – en somme un sujet de sociolinguistique qui merite certainement toute notre attention. [French]


INTRODUCTION
The term 'global linguistic variants' (e.g. Buchstaller 2008) has been coined for linguistic innovations arising contemporaneously in highly disparate places. The use of 'global' here is not intended to imply usage absolutely everywhereor even necessarily in different languagesbut just to highlight the sheer distances involved, and partly to intimate a possible role for globalised mass media. However, there remains deep division in sociolinguistics about that possible role. Debate has concerned, firstly, whether these reported global changes are any more than linguistically superficial, with negligible impact on deeper linguistic structure; and secondly, whether and how far the media might be involved. As Eckert judiciously puts it: We have all been told by non-linguist acquaintances that language change comes from the television. The idea that language change could be accomplished in such a trivial fashion is part of the popular 'bag o' words' view of language . . . that we're all tired of dealing with. However, we shouldn't ignore the possibility that not all changes are equal. We need to ask ourselves what kinds of changes require the kind of repeated exposure that regular social interaction gives, and what kinds can be taken right off the shelf. (Eckert 2003: 395) On the question of linguistic superficiality, Meyerhoff (1991) and Trudgill (2003) discuss the use of putatively U.S.-originating words in New Zealand, like flashlight for torch, or truck for lorry. Trudgill (2003) contrasts this lexical convergence with the steadfast divergence of New Zealand and U.S. vowel systems, to evince the impervious localness of these more complex systemic changes. Similarly, Milroy (2007, after Eckert as above) contrasts 'off the shelf' changes in the U.K.such as nationwide increases in 'vocalization of /l/ and glottalization of /t/ ' (2007: 164)with what she terms 'under the counter' changes, which require prolonged contact in social networkssuch as systemic vowel shifts. Relatedly, Dion and Poplack argue: 'although . . . "off-the-shelf" changes . . . can diffuse easily . . . the transmission of complex linguistic variables . . . requires face-to-face contact ' (2007: 1;cf. Barbieri 2009: 71). The matter of media influence has been less contentious with superficial changesthe 'diffusion of catch-phrases', as Chambers puts it (1993: 138-140; see also Charkova 2007;Risager 2006: 96-97). The debate warms up when apparently complex innovations are reported contemporaneously across great distances.
One global variant has attracted the most research: quotative be like (e.g. 'I was like, no way!'), displacing quotatives such as say across the Englishspeaking world. Meanwhile other changes, though not global, have arisen with similar speed across whole countries, such as TH-fronting to [f]/[v] in the U.K. Milroy (2007) includes both be like and TH-fronting in the 'off the shelf' category, suggesting a certain superficiality; but quotatives in particular have been argued to be complex, core grammatical, 'high-context' features (Buchstaller 2006a), part of a 'quotative system' (Tagliamonte and D'Arcy 2004) indicative of systemic change. As such, one might expect these to act as under-the-counter changes reliant on face-to-face contact, yet they are reported in 'discontinuous geographic settings' worldwide (Buchstaller and D'Arcy 2009: 291;cf. 'spatially non-contiguous communities', Buchstaller 2006a: 363). Could all these changes really be driven by contact among itinerant globetrotters? Their speed and reach makes it unclear; the scale seems out of proportion. The quandary deepens in the way such changes tend to be pioneered by adolescents, who lead very local lives yet have strong access to mass media (Altendorf 2003: 148-149). Milroy (2007) does end by tentatively questioning the explanatory adequacy of face-to-face contact, but she seems cautiously silent on the media specificallyonly highlighting a dearth of research into the social mechanisms underlying off-the-shelf changes. Her caution speaks to the level of contention at play on this issue. Where to go from here, and whether to invoke the media, has become one of the hottest potatoes in early twenty-first century sociolinguistics.
As we will see, research focusing on global innovations has made some claims about a role for the media, though these claims have tended to outpace the evidence. Language corpora have remained the datasets of choice, and that elides two important factors: firstly, the way global variants are actually used in mass media; and secondly, the way individual people engage with mass mediaand precisely how that might figure in their appropriation of variants. All this has led to ambiguity as to whether sociolinguistics can really account for global variants, and if so, what the new epistemological landscape might look like.
In this article I review recent research on global linguistic variants, together with two concepts that help to fill the above two gaps. First is mediation, a term developed in a different area of sociolinguistics to describe the way nonstandard vernaculars are (re)produced in media texts (Coupland 2009). Second is parasocial interaction (Rubin, Perse and Powell 1985), developed in media audience research to describe how media engagement works as a creative, emotionally involved process, actively reworking media content. As Livingstone notes of soap operas, they 'thrive on interplay between . . . viewpoints', thereby 'providing a considerable role for the viewer ' (1998: 42-52). As we will see, such emotional investment is central to any substantive role for the media in language change.
I use these foundations to develop an epistemological model for research on the media and language change: the mediated innovation model. Along the way, I arrange existing studies of global linguistic variants into five identifiable approaches, each one engaging more substantively with media texts and media engagement. All this is intended to clarify the achievements and limitations of current research, and to map out where future research could focus its efforts.
There will be some limitations in what follows. The field of research I am reviewing is at quite an early stage, with some critical blind spots. In the last of the five approaches proposed below, I describe a study by a team in Glasgow (Scotland), which goes beyond conventional methods and collects new kinds of data on media influence. Yet, this study focuses on innovations originating in England, meaning a far smaller spatial scope than the transatlantic cases reported elsewhere, let alone the global ones. The new data in the Glasgow study do indicate a significant role for the media, but face-to-face contact also weighs in quite heavilymore than one might expect had the same techniques been applied across global distances. So the overall case for media influence could be sharper than it is presently. But then, my purpose is not to propound media influence, only to build an epistemological model for its interrogation.
In any event, the studies reviewed below suggest that local social conditions ultimately decide the linguistic and demographic profile of innovations, and that any role for the media is much diluted by the 'repeated exposure' of 'regular social interaction' (Eckert 2003: 395). This in turn reflects audience research on the partial, negotiated role of the media in influencing behaviours: [L]ocal realities can . . . present an unpredictable interpretive screen through which the intruding electronic screen images are filtered. . . . Media reality has not completely erased social reality, . . . counterposed as it is by the centrifugal forces of the local micro-circumstances in which people live out their everyday lives . . .. (Ang 1996: 151-152;cf. Stuart-Smith 2011) A second limitation is a lack of clarity to date on whether global variants represent lasting changes, or merely fads (Barbieri 2007: 29;Tagliamonte and D'Arcy 2007: 213). As Mair puts it dryly, 'generations of teenage linguistic rebellion will not lead to a lasting change in community norms ' (2006: 29). Longevity is not altogether critical in this article, but it could be interesting in future to analyse longevity using the mediated innovation model as a backdrop.
Thirdly, my discussion involves only TV and film, not the internet. I am limited here by the scope of existing research. Despite interest in globalisation from the wider discipline of sociolinguistics (e.g. Blommaert 2010: 49), and emerging work on computer-mediated communication (e.g. issue 10:4 of this journal), geographically-minded variationists so far have not gone 'online'. There have been some variationist studies of email (Kuzmack 2010) and newsgroups (Buchstaller et al. 2010;Coppen and Foolen 2012: 271), but these have not tracked the geographical spread of innovations. Still, TV is not a moribund topic; global audiences continue to grow (BBC Worldwide 2011: 26). There is just scope for expansion in future research. I return to that at the end.
A final caution: the studies I review here mostly concern native Anglophone societies. Relevance to other contexts remains to be exploredsee e.g. Baird (2001: 7), Cukor-Avila (2002: 3), and Kohn and Franz (2009: 262). Perhaps that is for the best though, given the intended applicability to future research.
A further contribution of this article is to examine the way researchers construct their arguments in relation to media influence, including the use of rhetorical devices to bolster claims when empirical data may be lacking. This is itself an important sociolinguistic topic; it tells us a lot about the evolution of this field. I do not introduce rhetorical terms separately here, but as they arise.

KEY TERMS
Some terminological foundations need to be clarified before beginning the main discussion. I review five key terms here, to be used in what follows.

Sociolinguistic convention
For my purposes, this relates to a prevailing focus in variationist studies on face-to-face contact as a principal factor in language change. Neither theory nor methodology has delved far into media influence. Moreover, in the wider social science of globalisation, sociolinguistics has been 'late getting to the party' (Coupland 2003: 465). This point of departure is elaborated on below.

Diffusion of innovations
The spread of linguistic innovations across social and geographical space has been labelled 'diffusion'leading on, somewhat circuitously, from earlier behavioural studies (see seminal work by e.g. Rogers 1962). Labov (2007) reflects on his own and others' sociolinguistic research in this area. Linguistic 'transmission' is the 'unbroken sequence of native-language acquisition by children ' (2007: 346), enabling continuity of dialect norms. Such norms change by 'incrementation', starting with 'faithful transmission of the adult system, including variable elements' to children, followed by advancement of that variation (2007: 346). 'This is the normal type of internal language change, termed CHANGE FROM BELOW or change from within the system ' (2007: 346;cf. Macaulay 2006: 280). This is contrasted with 'CHANGE FROM ABOVE or the importation of elements from other systems' (Labov 2007: 346), involving 'learning, primarily by adults, who acquire the new variants of the originating community' and then transmit them to the next generation back in their community (2007: 380)designated as 'linguistic DIFFUSION' (2007: 347). Britain (2005: 996-997) compares types of diffusion: • wave or contagion, where an innovation spreads out in a 'ripple'; • urban hierarchical, from large cities to towns, villages, and so on; • contra-hierarchical, from rural to urban areas; and • cultural hearth diffusion, the innovation taking hold simultaneously in both urban and rural areas before spreading elsewheresee Britain (2013) for developments and critiques of these models.
A crucial point of departure below is that all these types of diffusion are premised on face-to-face contact.

The speech community
This term was developed by Bloomfield (1926: 153-154) to refer to the totality of alike utterances (acts of speech) in a given community; he gives Frenchmen and Zulus as examples. Subsequent reworkings of the term from the 1960s onwards, by the likes of J. J. Gumperz and Dell Hymes, zoomed in further than discrete languages, emphasising local variation. Reviewing a broad range of twentieth century sociolinguistic studies, Patrick (2002: 577) arrives at a view of the speech community as a 'socially-based unit of linguistic analysis', realised by shared productive and evaluative language norms, and usually some chain of interaction linking its members (cf. Owens 1999: 663).
It is not necessary for all these members to actually meet, so some speech communities have grown very large. Most members of the 'African American speech community' (Rickford 1999) never meet, but racial divides in the U.S. have helped reinforce the distinctions of African American Vernacular English (AAVE) (Fridland 2003). A relatable speech community comes in the form of a Mexican enclave of a U.S. city, with an 'emerging Hispanic English' under little influence from ambient southern U.S. speech communities (Wolfram, Carter and Moriello 2004). Despite this local isolation, there is a flow of migration into the enclave from faraway Mexico. This speech community is locally discrete, yet it has one arm stretched out hundreds of miles south-west, with productive and evaluative norms following a discernible chain of interlocution.
Diffusion serves to connect otherwise discrete speech communities, usually via adults traveling between them. Milroy notes that 'off the shelf' changes ' [do] not require the support of . . . regular primary interaction and [are] therefore more generally accessible to mobile or marginal individuals ', like commuters (2007: 163), but less accessible to people with low mobility and denser social network ties: 'Because of the . . . time and commitment needed to maintain these strong ties, they . . . lack opportunities to form . . . weak [external] ties' (Milroy and Milroy 1985: 367-368). Crucially though, although diffusion involves interaction beyond the speech community, it still relies on contact between interlocutors. A possible role for the media goes beyond that. To return to AAVE, consider a related music genre, hip-hopalso widely appropriated globally (e.g. Cutler 1999; Lee 2011). AAVE hip-hop artists in the U.S. are, at one level, all interacting members of the AAVE speech community. However, when they become superstars, their interlocution becomes heavily imbalanced. Their words are heard by far more people than could ever speak back to them, and who will never meet them. This is more than diffusion of off-the-shelf changes via mobile individuals: it is not limited to face-to-face contact; it follows no identifiable geographical pattern; and access to these artists' linguistic output has no link to the strength or weakness (or even existence) of social network ties. If mass media have a distinct role in spreading innovations between distant speech communities, then it would be a codicil to a model based principally on face-to-face contact and network ties.

Globalisation
It is often contested that globalisation is particularly recent. Taking the U.K. as an example, consider a warning from A. J. Ellis writing in the nineteenth century: 'the present facilities of communication are rapidly destroying all traces of our older dialectic English' (1999[1871]: vi). So, what's new? For one thing, the changes bothering Ellis were limited to 'dedialectalisation' (Trudgill 2002: 40)encroachment of Standard English on to local dialect forms. I have argued elsewhere (Sayers 2009: 51-58) that those changes, dominant in the U.K. from the late-nineteenth to mid-twentieth century, were due to increased literacy, and the commensurate influence of written Standard English (cf. Chambers 1993: 141-142). This period saw little transfer of non-standard dialect forms between even contiguous speech communities, let alone distant ones.
Geographical space can be defined 'in terms of relations' between people (Allen, Massey and Cochrane 1998), which greatly increased in scope in midlate-twentieth century Britain, alongside rising geographical mobility. In this period, dedialectalisation was gradually eclipsed by dialect contact and mixture (Sayers 2009: 59-155), allowing off-the-shelf changes to accelerate apace. This was a change from Ellis' day, but did it relate to globalisation? Scholte (2005: 59-84) argues that globalisation only began in earnest in the 1980s. He stresses the quantitative scale of 'transworld associations, communications . . ., finance, investment, travel and trade', alongside the qualitative distinctness of 'social connections that substantially transcend territorial geography ' (2005: 61). 'The newer spread of transplanetary simultaneity and instantaneity takes social relations substantially beyond territorial space' (2005: 62, emphasis original). Scholte includes 'global mass media spread[ing] messages simultaneously to transworld audiences ' (2005: 62) in his examples. So, increased mobility and dialect mixture was a change from Ellis' day, but it bore no hallmarks of globalisation per se (cf. Buchstaller 2008: 19). Peculiar to globalisation would be the co-occurrence of innovations in discontinuous geographic settings, inexplicable solely on grounds of mobility and personal contact. This is precisely what has motivated discussion of media influence in the rise of global linguistic variants. Moreover, the aforementioned sociolinguistic concept of mediation is centrally associated with the 'uncoupling of space and time' in the conditions of globalisation (Fairclough 2006: 85, after Thompson 1995.

The influence of mass media
Altendorf offers a frank reflexivity about media influence on language, in a discussion of innovations co-occurring across the U.K., worth citing at length: We have already identified these variants as 'youth norms' . . . and not as south-eastern [U.K.] accent features . . .. We have not been able to explain how they could have spread without extensive face-to-face interaction, which is generally regarded as a necessary pre-requisite for diffusion . . ..
It is possible but not yet proven that the combined effect of linguistic preference + social attractiveness + constant exposure through the media might have an effect . . ..
In any case, the development has reached a point where we might have to re-evaluate the role of television in bringing about new linguistic trends. . . .
[F]or example, . . . if TV watching trends continue, 'many children will come to spend more time exposed to non-local varieties than to their local vernacular' (Foulkes and Docherty 1999: 15). (Altendorf 2003: 148-149) Foulkes and Docherty are similarly circumspect: 'The possible effects of [media] exposure on acquired speech patterns remain to be formally tested, but are certainly worth considering in the context of continuing changes' (1999: 15). Kerswill and Williams (2000: 105) describe T-glottaling (on the rise across the U.K.) as 'particularly evident in the spoken media, to which children are increasingly oriented'. Trudgill (1988: 43) notes of U.K.-wide TH-fronting: What is surprising . . . is the extreme rapidity of this change. . . . The pattern of geographical diffusion suggests very strongly that face-to-face contact, as a result of mobility and immigration . . . must be involved (see Trudgill 1986). The sheer speed of the change, however, may be due to a softeningup process . . . through television programmes . . . .
The media are given a rather minimised role here; the 'softening-up process' seems to mean only greasing the wheels for changes occurring anyway via conventional diffusion. This minimisation is not about fingers jammed in ears, just reasoned scepticism that diffusion cannot ultimately explain everything. The studies of global innovations reviewed below have departed from this line.

FIVE APPROACHES TO GLOBAL INNOVATIONS
I identify five approaches in existing research on global innovations. Each embodies a further departure from sociolinguistic convention. These five are outlined briefly below, and then each is expanded in its own subsection.
Approach 1 comprises studies on innovations conducted in just one speech community, with only tentative mentions of globalisation and media. This is the most local approach, with the highest constraints on any 'global' argument. In Approach 2, innovation usage is compared in discontinuous geographic settings worldwide. This is a clearer nod to the global, but only by virtue of distance and conspicuous linguistic similarity; there is still little mention of the media, or other unconventional factors. In Approach 3, the role of the media is asserted more explicitly, yet still not explored empirically. In Approach 4, innovation usage is recorded in media texts as well as speech corpora. This achieves a methodological advance by analysing mediation, but it does not explore how people actually engage with mass media. Approach 5 progresses beyond the methodological conservatism of Approaches 1-4 with a tripartite exploration of conventional speech data, mediation in media texts, andmost importantlymedia engagement practices. Crossing this Rubicon enables a significantly fuller interrogation of media influence.
Note that although Approach 1 cites the oldest studies, and Approach 5 the newest, overall they are not chronologically ordered. They do not represent a shift in empirical claims over time, just different kinds of argument.
Approach 1: One speech community at a time Blyth, Recktenwald and Wang (1990) examine be like among respondents at Cornell University, according to age and gender. Ferrara and Bell, at a university in Texas, analyse a corpus recorded by three student cohorts, and the 'gender-marking, age-grading, ethnic distribution, and rural-versus-urban usage' of be like (1995: 270). Neither study concerns the way be like arose in Table 1: Five approaches to the study of global linguistic variants, each representing a further departure from sociolinguistic convention *Not in all studies cited under this approach. these places. Ferrara and Bell note that be like is 'readily observable ' among TV entertainers (1995: 288), but this is just a brief aside in a footnote. Baird (2001) takes a similar approach to quotative usage in New Zealand, analysing age, sex, and other factors. D'Arcy (2010), also in New Zealand, and Cukor-Avila (2002) in a rural Texan location, compare usage by ethnicity. Baird briefly mentions the media near the end, then states: 'How much a trend catches on might depend on our (subconscious) attitude toward, and . . . exposure to, the nation responsible for starting it, in this case, the United States ' (2001: 18). This is the penultimate sentence of the article though, almost an afterthought. D'Arcy (2004, 2007) analyse realisation of quotatives in Toronto. They give a brief introductory review of localisation patterns in the U.S. and U.K. (2004: 509-511), but focus mainly on the Toronto data. They discuss why be like arose in Toronto after its earlier rise in California: 'As part of the "preppie" movement of the 1980s, be like gained prestige as a trendy and socially desirable way to voice a speaker's inner experience ' (2007: 212). Again though, this is quite a casual assertion, right before their conclusion. Detail of how be like might travel is not pursued.
Macaulay (2001) reports the pioneering of be like by adolescent girls in Glasgow. He asserts its origin in California (2001: 3), and queries the primacy of conventional diffusion: 'the question remains of how be like reached teenagers in Glasgow. It is unlikely to have been through direct contact with young Americans ' (2001: 17). He cites a study showing increased use of be like in American films over the previous two decades, adding: 'it is possible that the innovation . . . owes something to the media ' (2001: 17). This is a swift remark though, just before the conclusion which simply recommends '[m]ore attention to archival media materials ' (2001: 18) to explore such a link. Be like is noted as originating peculiarly far away, and some initial interest in the media is clear, but Macaulay's attention mostly remains on usage patterns of be like in this one location, not the role of the media in its arrival there. Kohn and Franz (2009) analyse ethnicity and usage of global variants across North Carolina. Early on they lament the lack of research on 'the impact various social and local identities may have on the distribution of a global form'. But they go on to focus tightly on linguistic data, only returning in the conclusion to a brief and fairly abstract discussion of 'structural differences', such as 'access to telecommunications . . . or transportation ' (2009: 285). Meyerhoff and Niedzielski (2003), in New Zealand, analyse the use of innovations considered to have U.K. and U.S. origins, and whether 'instead of being seen as borrowings, [they] are perceived to be home-grown variants ' (2003: 549). They consider 'some general principles . . . in the social sciences . . . associated with globalisation and . . . the theoretical and methodological implications . . . for the study of language variation' (2003: 534). They provide a novel discussion of 'tensions between globalisation and localisation ' (2003: 550). Overall though, they remain cautious about reasons for this cooccurrence, and focus mainly on local attitudes to the innovations. Just before their conclusion, they cite some studies on mass media, and set a future goal for sociolinguists to explore how 'patterns of language use reify or challenge the interests of globalisation ' (2003: 549). In all, they offer quite a lengthy discussion of globalisation, but they do not directly claim media influence, nor do they compare corpora from discontinuous geographic settings.
Stuart-Smith, Timmins and Tweedie (2007), in Glasgow, examine TH-and DH-fronting, L-vocalisation, R-labialisation, and T-glottalling. Some of these were already familiar in the local dialect, but they were arising in unfamiliar phonetic environments, leading to 'a consonantal system which in many respects is more similar to that of London English ' (2007: 222). As noted earlier, although this is only between Scotland and England, the authors raise similar queries to those researching global co-occurrences. Stuart-Smith, Timmins and Tweedie (2007: 250) report that middle-class adolescents 'largely maintain Scottish regional standard norms' and show no TH-or DHfronting, low L-vocalisation, and only moderate T-glottalling; middle-class adults maintain standard norms even more so; and while working-class adults are losing some local dialect forms, it is working-class adolescents who are the 'leaders of change ' (2007: 251). The authors see this as a very normal attempt to sound 'as anti-middle-class, and anti-establishment as possible ' (2007: 251); the only question is how these 'less-mobile, strongly-tied, working-class adolescents ' (2007: 224) came to pioneer these non-local variants. The authors discuss the relative placelessness of these innovationson the rise among young people across the U.K. and apparently existing 'in "cultural" or ideological space, towards which speakers may orientate ' (2007: 224). Diffusion in social networks ultimately does explain much of their data, but they end with three 'troublesome "non-local" variants, [f], [v] and vocalised /l/ ' (2007: 254), which evade such explanation. 'Our information for workingclass adolescents reduces the likelihood of dialect contact as a direct factor ' (2007: 255). The authors do not probe media influence, and are careful not to speculate, although they mention (2007: 224)  Descriptively they are using a mixed consonantal system, with local and non-local features. Whether they intend this repertoire to sound mixed, or anything other than 'pure Glaswegian', seems unlikely, though that in itself does not rule out interaction with television or the media as additional contributory factors in these changes. But that is another story altogether.
Broadly, Approach 1 involves analysis of interesting innovative forms, but not cross-corpus comparisons of discontinuous geographic settings, nor much discussion of how the innovations got from place to place. Some studies briefly mention the media, but only cautiously in passing. Approach 1 is represented in Figure 1, illustrated with data from Macaulay (2001). In Figure 1 the outer oval represents the Glasgow speech community; the inner ovals represent the groups of speakers categorised by Macaulay and their innovation usage rates. Epistemologically, Approach 1 looks a lot like any other variationist study; the only difference is how the authors puzzle over the way the variants arrived.
Approach 2: Compare discontinuous geographic settings Winter (2002), building on Tagliamonte and Hudson (1999) and Ferrara and Bell (1995), compares quotative usage in corpora from Australia, the U.K., and Canada. Comparing corpora from distant speech communities helps to substantiate the spatial scope of the co-occurrence. It also elucidates the localisation of global variants, the unique sociolinguistic profiles they develop in each place. Still, the analysis remains focused on linguistic detail. There is little interest in how this global co-occurrence came about. At one point, Winter recalls an Australian TV character's 'stereotypical', 'exaggerated usage' of the kinds of innovative quotatives found in the Australian English corpus (2002: 12). This is an isolated aside though, developed no further. Barbieri analyses quotative use by age and sex in a corpus recorded 'across numerous US states ' (2007: 31). Barbieri (2009) elaborates by comparing two pan-U.S. corpora recorded a decade apart, noting sustained growth of be like. As for underlying mechanisms, she offers macro-level interpretations based on other attitude studies (2007: 41), but specific mechanisms are not pinpointed; nor are data collected from individual speakers about influences on their usage. Durham et al. (2012) compare usage of be like in two cohorts of undergraduates in York (England) a decade apart. They also provide attitude data on say and be like in different linguistic contexts from a cohort of American English speakers. They very briefly assert that 'be like has diffused geographically' (2012: 328) and 'continues to diffuse globally' (2012: 327), but they otherwise do not explore mechanisms behind this distant co-occurrence. Tagliamonte and Hudson (1999) are somewhat bolder. Comparing corpora from York (England) and Ottowa (Canada) to Ferrara and Bell's (1995) Texas data, they assert the 'remarkably parallel' usage rates as 'evidence for a systematic global diffusion of be like across geographically separated speech communities ' (1999: 147), 'a very good linguistic indicator of . . . developments and changes we might expect from the putative ongoing globalization of English ' (1999: 168). But this venturesome claim is in their final paragraph, a pensive postscript where they also stress: 'the social mechanism(s) underlying these processes are beyond the scope of the present investigation ' (1999: 168).
Approach 2 maintains a largely conventional methodology of comparing speech corpora, just ones recorded in highly distant speech communities. Apart from some tentative, isolated asides in the sunset of their arguments, these studies give a sense of either trepidation or indifference towards the media.

Approach 3: Pinpointing the media
In Approach 3, globalisation and mass media begin to be discussed more explicitly, yet still within familiar empirical territory. A rift begins to open here between evidence and assertions, bridged in some cases by rhetorical devices.
Buchstaller compares U.S. and U.K. corpora to show linguistic, social and attitudinal adaptation of innovations, amounting to changes in 'perceptual load' (Buchstaller 2006a(Buchstaller : 363, 2006b. Buchstaller (2008) discusses globalisation at length, even distinguishing the role of geographical mobility and media. She argues that innovations spread globally 'in all probability mainly via the media' (2008: 37). Ultimately though, out of the 'global spread of innovations and their localized adaptation ' (2008: 21), only the latter is supported by evidence.
Buchstaller and D'Arcy compare linguistic localisation of be like in U.S., U.K., and New Zealand corpora: the unique way be like is used in each discontinuous geographic setting. They conclude that be like has undergone 'weak transfer ' (2009: 316), with most of its functions decided by local sociolinguistic processes after its global journey. They generally focus on linguistic detail, but make repeated gestures to globalisation, and claim the 'global transfer', 'global spread ' (2009: 316) and 'global diffusion' (2009: 317) of be like, describing it as one example of 'linguistic innovations that float globally through real (or cyber) space ' (2009: 293-294;cf. Buchstaller 2008: 19). They assert that be like has 'spread to numerous varieties worldwide in a very short time span through presumably limited interpersonal contact ' (2009: 322). They also raise the quandary that be like is pioneered by low-mobility adolescents with strong access to 'meditated forms of communication (e.g. email, IM, blogs, wikis, TV, etc.) ' (2009: 322). They foreground globalisation and media fairly boldly, but the discussions of 'global spread' drift free of the empirical support that underlay their main linguistic analysis. This amounts to the use of a rhetorical device, proof surrogate, where media influence is inferred but not evidenced. Buchstaller (2004) states: 'The very social group which introduces like into the linguistic system is also the one we can assume to be most likely to pick up a new lexical variant through soaps, talkshows, etc. ' (2004: 289). There follows a countervailing emphasis on the role of face-to-face contact, to stress 'that mass media are an insufficient transmission channel for the whole variant (surface item, functional value and social value) ' (2004: 290). At first blush, this appears to echo media audience research on the partial, locally contested influence of media stimuli (e.g. Liebes and Katz 1990), but that echo fades at a critical juncture. Unlike audience research, there are no data from participants about their media engagement practices, only the speech corpora. Given that empirical disjuncture, Buchstaller is here deploying a different rhetorical device, namely paralipsis: the media cannot do everything, but this is asserted precisely to suggest they are doing somethingyet without evidence for that something.
In Approach 3, the methodologies remain largely unchanged, but the claims of media influence grow. As a result, gaps open up in the empirical defence. Overall, Approaches 1-3 do not explore how 'globally travelling features' (Buchstaller 2006a: 375) actually travel.
Approach 4: Comparing speech data and media data Tagliamonte and Roberts (2005) analyse intensifiers (e.g. very, so) in the U.S. sitcom Friends, and compare speech data from the U.K. and North America. Linguistic detail dominates their article, in which correlations arise: 'The Friends data exhibit almost the same overall rate . . . as . . . contemporary English'; 'media language actually does reflect what is going on' (2005: 296). They infer media influence -'these media data appear to pave the way; language is more innovative in the media than in the general population' (2005: 296)but end inconclusively on that point: 'At the very least, the inextricable link between language and societyoften self-evident for sociolinguists and dialectologistsstands out in this icon of pop culture ' (2005: 297). They seem to inch up to, then step away from, an explicit assertion of media influence. Dion and Poplack (2007) make a similar comparison of speech data and media texts, although 'quotative use in the scripted media . . . diverged wildly from the community norm' in their data, so they remain doubtful of influence.
Approach 4 answers Macaulay's appeal for '[m]ore attention to archival media materials ' (2001: 18). Still, Approaches 1-4 do not probe how individual speakers' media engagement practices might contour their use of innovations. This is touched on, but only as an addendum to the main linguistic analysis. Amid discussion of the 'globalisation' and 'localisation' of variants, it is the latter to which such corpus analyses are limited. Precedence does not entail ancestry, and assertions of global 'spread', 'diffusion', 'transfer', etc. are strictly post hoc, assuming cause and effect from temporal successioncf. Britain's (2013: 475) distinction between portrayal and explanation of sociolinguistic patterns. As Buchstaller notes: 'There is still a dearth of detailed micro-linguistic analyses that demonstrate the links between local linguistic mechanisms and global forces ' (2008: 20). That dearth is addressed in Approach 5.
Approach 5: Gauging the effect of the media Media engagement as a factor in language change is empirically substantiated in only one study, by a team led by Jane Stuart-Smith in Glasgow. This was a follow-up to their earlier study in Glasgow, included in Approach 1 above. That earlier study failed to explain its data with social networks alone, but lacked the means to explore media influence. The consonantal changes of THfronting, DH-fronting, and L-vocalisation are seen by Milroy as 'off the shelf' changes, 'accessible to mobile or marginal individuals ' (2007: 163). But, as Altendorf notes, these are 'youth norms ' (2003: 148-149). As the Glasgow team found, the quandary for contact-based explanations was the way these changes were being led by adolescents, typically with the lowest mobility. Trudgill includes TH-fronting in Glasgow in a list of 'extremely similar phonological changes . . . taking place in varieties of English around the world more or less simultaneously', which evade 'explanation in terms of diffusion ' (2003: 56). Ever sceptical of the media, he argues that these changes are in fact independently driven by predetermined structural linguistic trajectories. Meyerhoff and Niedzielski (2003: 546) set a similarly high bar, stating that 'the possibility of independent, parallel development' must first be ruled out. A further possible explanation for the Glasgow data could have been adolescents adopting variants from geographically mobile adults, then simply using those variants more. With the new data in their follow-up study, Stuart-Smith and her team found that such conventional influences did play a part, but only alongside, and indivisibly from, an empirically substantiated role for the media.
To collect their new data, an interdisciplinary methodology was developed with academic collaborators specialising in audience research and statisticssee Stuart-Smith et al. (2013). First, London-based TV shows were identified by the participants. That decisive role for the participants was an advance from e.g. Tagliamonte and Roberts (2005), giving a firmer basis for subsequent claims of influence on those participants. The researchers then recorded rates of the above consonantal variables in these media texts, to operationalise 'media-Cockney' as a source variety. As discussed at the outset, this step illustrates the process of mediation.
To explore media engagement (as well as other conventional sociolinguistic factors), speech data were combined with participant observation, interviews and questionnaires over a four-month period in 2003-2004. This was geared towards fine-grained mapping of social networks, and specific examination of how media engagement correlated with and contoured innovation usage. Following the measurement of innovation usage in the Glasgow data, the first attempt to account for these patterns was a large-scale regression analysis of [f] and [v] . . . and numerous extra-linguistic factors . . .: potential for dialect contact within and beyond Glasgow, attitudes to urban accents, social practices and identity, entertainment preferences, engagement with computers and the internet, involvement in sport, and exposure to and engagement with television, both generally and with specific programmes. (Stuart-Smith and Timmins 2009: 42) The analysis showed multiple influences: 'limited contact between weakly tied individuals, . . . positive attitudes towards London accents (Milroy 1987: 203;Trudgill 1988), and/or orientation towards 'youth norms' experienced partially via . . . broadcast media Kerswill 1999)' (Stuart-Smith andTimmins 2009: 41). 'Adopter categories' were used to distinguish participants by their level of innovation use; these included 'innovators' and 'early adopters' cf. Milroy and Milroy 1985: 367). Regressions of the linguistic and extra-linguistic data showed that innovation use correlated with local contact among adopter categories, media engagement, and contact with people from the London-southeast region (Stuart-Smith and Timmins 2009). However, each extra-linguistic factor alone gave a poor overall explanation of innovation use; 'the final models with the full range of social factors' explained innovation use 'at least three times better than any single category model alone ' (2009: 42). Recalling the explanatory ponderings in Approaches 1-4 between media and face-to-face contact, here is the first empirical illumination of that balance. Different factors coincide, overlap, and interrelate; their effects cannot be analysed or asserted discretely.
At the level of individual participants, influences varied. For some, contact with people outside Glasgow emerged as the strongest factor (Stuart-Smith and Timmins 2009: 52). For others, media influence had a firmer correlation. The respondent with by far the most DH-fronting had no interpersonal contact outside Glasgownot even much insideyet very high engagement with the London soap opera EastEnders (Stuart-Smith and Timmins 2007: 20). Media engagement functioned as a form of parasocial interaction; indeed it is those with limited social contact for whom Rubin, Perse and Powell (1985) argue that parasocial interaction is the most potent. More generally, there was a robustly significant relationship between innovation use and prolonged, emotionally-invested media engagement (Stuart-Smith 2012). Such participants were innovators and early adopters (Stuart-Smith and Timmins 2009: 46). Importantly though, prolonged exposure without emotional investment had no such significance. This reaffirms the importance of parasocial interaction for media influence.
All of this being said, engagement with London-based TV did not mean positive evaluation of Cockney accents. Similarly to the attitude studies cited earlier (e.g. Meyerhoff and Niedzielski 2003), London accents received mixed evaluations, while the innovations themselves were considered by participants as 'pure Glaswegian' (Stuart-Smith 2007: 12). Indeed, subsequent reanalysis of the data (Stuart-Smith et al. 2013) has further lessened the significance of attitudinal factors. These innovations in Glasgow gained a new 'perceptual load' (Buchstaller 2006a(Buchstaller : 363, 2006b. They did not keep their 'functional and social boots on' (Buchstaller and D'Arcy 2009: 282). The authors provide an illustrative vignette from their data collection process, when they were recording the young people reading word lists: the teenagers treated the task as an opportunity to display . . . 'their' speech, and one can hear them laughing and playing up to the microphone . . .. The adolescents are . . . representing their own group, or at least a possible version of their . . . repertoire . . .. What is interesting is that the non-local non-standard variants are selected for this particular stylistic repertoire. Keen to emphasise the complexity of overlapping influences, the authors point to historical evidence that these innovations were present at some level in Glasgow before their contemporary surgeperhaps due to historical contact with Londonersand that media engagement served only as an accelerating factor (Stuart-Smith and Timmins 2014). This supports, and augments somewhat, Trudgill's (1988: 43) 'softening-up process' from TV. Approach 5 may be the furthest departure from sociolinguistic convention, but Stuart-Smith and her team are no iconoclasts. Stuart-Smith considers their argument amenable to Trudgill's (pers. comm. 23 May 2012), indeed building upon the insights from his work in Norwich. As she notes, Trudgill had reported THfronting there among low-mobility, working-class speakers, driven partly by contact with Londoners, but he also speculates about the potential role of television programmes based in London in promoting positive attitudes towards London dialect features: 'television may be part of a "softening-up" process leading to the adoption of the merger [of /f/ with /th/], but it does not cause it' (Trudgill 1986: 55). (Stuart-Smith 2006: 141) Stuart-Smith and Timmins reject the idea of 'blanket transmission from media source to passive speaker/viewers, a notion long abandoned within media effects research ' (2009: 54). Overall, the picture that emerges is one of 'different causal pathways, and combinations of pathways, for different speakers' (Stuart-Smith and Timmins 2007: 28). The innovations in Glasgow arrive via crosscutting and interrelating routes, in which the media play a demonstrable part, but indivisibly from concurrent social processes.
Approach 5 progresses beyond inferences derived from macro-level theories (characteristic of Approaches 1-4), and answers Buchstaller's appeal for 'detailed micro-linguistic analyses that demonstrate the links between local linguistic mechanisms and global forces ' (2008: 20). As for the construction of their arguments, Stuart-Smith and her team are of course advancing claims based on inevitably limited data, but they strike a shrewder, more painstaking balance between evidence and assertions. Collecting actual data about media engagement practices has led, perhaps counter-intuitively, to a more subdued assertion of media influenceemphasising uncertainty and deferring to complexity. Truth is stranger than fiction, as it were.
The Glasgow study serves to query the idea that only weakly tied individuals introduce non-local variants. If there is a discernible role for the media, then it is not limited in this way, since there were innovators and early adopters in Glasgow with high, positive media engagement, strong local networks, and weak external ties (or none). This reprises an overarching point: if there is a role for mass media, it is qualitatively distinct from conventional diffusion.
Approach 5 involves a combination of different methodological tools for analysing media engagement alongside other factors. It is the epistemological mapping of these different methodologies to which I now turn.

THE MEDIATED INNOVATION MODEL
Following Coupland (2009), as cited at the outset, vernacular forms used in media texts can be said to have been mediated. If these mediated forms are adopted as innovations in discontinuous geographic settings, then I suggest the term mediated innovations. Figure 2 depicts the mediated innovation model, an epistemological model for visualising methodologies used to research media influence in language change.
The conventional processes of transmission and diffusion rely on contact among interlocutors. Diffusion occurs between speech communities via weakly tied mobile individuals. If there is a role for media engagement, which does not rely on personal contact and occurs regardless of tie strength, then I would refer to it as innovation broadcast. Mediation can be seen as a prior step to this. As the Glasgow study in Approach 5 shows, mediation and broadcast occur alongside, and intertwined with, conventional diffusion. It is precisely this rich mix of influences for which the mediated innovation model is designed. Figure 2 is a diagrammatic representation of a mediated innovation model. It depicts the datasets resulting from the various methodologies used to probe 'different causal pathways' (Stuart-Smith and Timmins 2007: 28). The ovals in the diagram represent language corpora, from distant speech communities and from media texts. The dotted backgrounds represent social networks through which diffusion occursdenser within speech communities, weaker between. The dots do not permeate either the media texts or the processes of mediation and broadcast. As argued so far, although these interrelate with conventional diffusion, they are qualitatively distinct, and require different empirical data. Figure 2 sacrifices detail for simplicity, which exposes certain risks. At a glance, it might even seem to suggest a highly deterministic role for 'media texts', downplaying the complexities of media engagement and localisation. Figure 2 could be redrawn in various ways, perhaps better in three dimensions, but the aim is only to set out clearly the different sites for analysis. It is not a theoretical model but an epistemological model, depicting the columns of Table 1 as a kind of checking exercise to see what has been empirically substantiated.
The model also inevitably falls back inelegantly on some sociolinguistic convention which it would have been ideal to progress beyond. The distinction of 'source' and 'adopting' speech communities may seem curiously linear in a discussion of global flows. That distinction emanates from the studies reviewed above, all of which compare innovation usage in a putative antecedent speech community and one or more adopting speech communities. Such distinctions may well be deconstructed in future research, in which case the model will hopefully adapt. 'Adopting' may also seem a little simplistic; it is intended here as shorthand for the multifarious adaptations, appropriations and localisations that incoming innovations go through. As Mair puts it: Particularly when it is nonstandard forms of American English which are spreading . . . we are rarely dealing with simple . . . linguistic Americanization but with a more complex . . . negotiation of vernacular norms in a globalized communicative habitat. (Mair 2006: 195) Let me walk through how the model helps to elucidate the limits of the studies reviewed above, with a view to how it can help structure future work. Macaulay's (2001) study, cited in Approach 1 and represented in Figure 1, would fit inside just one 'adopting speech community' oval in Figure 2. None of the rest of Figure 2 applies. The studies in Approach 2 compare innovation usage in discontinuous geographic settings, like the U.K. and Canada (Tagliamonte and Hudson 1999). In such studies, the two 'adopting speech community' ovals in Figure 2 are present. If usage in the putative source variety is also analysed (as in Buchstaller and D'Arcy 2009), then the 'source speech community' oval is present. Still, throughout Approach 2 the 'media texts' oval is absent, as are all the arrows; those methodologies are not in use. Much the same is true of Approach 3: despite explicit mention of globalisation and mass media, neither mediation nor broadcast is interrogated or evidenced.
Approach 4 makes a substantial methodological advance, comparing speech corpora to media textsfor example Tagliamonte and Roberts' (2005) comparison of Friends scripts to speech corpora from North America and the U.K. Here, both 'adopting speech community' ovals in Figure 2 are present, as are the 'media texts' oval and the 'mediation' arrow. The 'broadcast' arrows are absent though; no data are collected about media engagement practices.
Although the Glasgow study in Approach 5 does not fill Figure 2 entirely, it does have the means to do so. The 'source speech community' oval is present, given the identification of London English features. The 'media texts' oval and 'mediation' arrows are present, given the operationalisation of 'media-Cockney' in TV shows identified by the participants. The 'broadcast' arrows are realised by the exploration of media engagement practices. As it stands, only one 'adopting speech community' is present (Glasgow), but if the study had incorporated another in a discontinuous geographic setting, then Figure 2 could have been illustrated fully.
With all these applications of the mediated innovation model to the studies reviewed so far, my purpose is not to claim things are missing from those studies, but to clarify what can be reliably asserted with their data. In turn, the mediated innovation model can help to apply some foundational sociolinguistic concepts to mediated innovations. First of all, actuation: What factors can account for the actuation of changes? Why do changes in a structural feature take place in a particular language at a given time, but not in other languages with the same feature, or in the same language at other times? (Weinreich, Labov and Herzog 1968: 102) Milroy and Milroy (1985: 342-344, 364-366) regard Labovian diffusion as an aspect of actuation, insofar as it connects distinct speech communities via weakly tied mobile individuals. Mediation and broadcast similarly connect distinct speech communities, though in a qualitatively distinct way. Actuation can therefore be expanded to accommodate mediation and broadcast as well as diffusion. Like diffusion, broadcast will vary in its intensity in different places, and may not happen at all in someas reflected in the varied localisations of mediated innovations, noted in the studies reviewed above. The mediated innovation model helps arrange the analysis of these localisations.
All this gives a new slant to Kerswill and Williams' (2000: 92) comment: 'Essentially, a close-knit network will resist . . . changes, unless these changes come via an "insider" who also has "weak ties" elsewhere (Milroy and Milroy 1992)'. In its original sense, this comment relates to Labov's 'CHANGE FROM ABOVE or the importation of elements from other systems' (2007: 346) as cited earlier actuated by weakly tied mobile individuals. But if there is a definable role for the media, then actuation may happen regardless of tie strength, and media input may constitute a quite different kind of link to 'outside'.
Next, embedding: 'How are the observed changes embedded in the matrix of linguistic and extralinguistic concomitants of the form in question?' (Weinreich, Labov and Herzog 1968: 101, emphasis original). For mediated innovations, embedding relates to the distinct local realisations in each adopting speech community. The mediated innovation model helps map out these local idiosyncrasies and the methodologies that interrogate them.
Mediation and broadcast, like diffusion, can connect distant speech communities. Unlike diffusion though, they are not limited by the extent of contact between weakly tied mobile individuals. The mediated innovation model helps to discern these qualitatively distinct and diverse links.
For clarity, a contrast can be made to the 'linguistic community'. Silverstein (1996) defines this as a group of disparate speech communities sharing an external target varietyeither a standard language, or a less formal orientation to another target variety. Mediated innovations do not depend on recognition of a target variety, as evinced by the recurrent perception that they are in fact local (Meyerhoff and Niedzielski 2003;Stuart-Smith 2007: 12). It is also worth drawing a brief contrast to 'imagined communities' (Anderson 1983); this idea describes the importance of print media to nationalist sentiment, not the influence of media (printed or otherwise) on sociolinguistic variation.
To sum up then, what use is the mediated innovation model? As an epistemological model, its main purpose is to arrange the methodologies used to research mediated innovations. Those methodologies account for previously unaccounted-for findings, namely the pioneering of non-local innovations by strongly-tied, low-mobility adolescents. The mediated innovation model brings in mediation from a different area of sociolinguistics, and specifies a further sociolinguistic process for investigation: broadcast. The model also distinguishes mediation and broadcast from conventional transmission and diffusion. All this helps to draw apart the different kinds of data and methodologies relevant to exploring these qualitatively distinct and concurrent processes. The model also throws into sharper relief the limitations in existing research, and clarifies the remit of future work. The model itself, and the emerging body of research it describes, will help sociolinguistics contribute to the social-scientific study of globalisation.

CONCLUSION
This fact of different places becoming swept up in similar economic, political and cultural flows . . . does not necessarily mean that places worldwide are becoming more alike. . . . '[N]on-local' processes combine with existing local differences to yield unique outcomes. Places 'internalize' these processes in distinctive ways, which is why place interconnection does not imply increased homogeneity among places. (Castree et al. 2004: 68, emphases original) Milroy (2007: 163) argues that 'socially embedded sound changes should be distinguished with respect to their different underlying sociologies and social trajectories'. For mediated innovations, I have argued that Approach 5 is currently the most advanced distinction of these sociologies and trajectories. The mediated innovation model offers perspective for whatever partial, negotiated role the media might play in the complex pathways that innovations take between discontinuous geographic settings, in conditions of globalisation.
[T]hough even our terminology for . . . 'media influence on language' . . . implies . . . the media somehow doing things to the viewer, it seems much more likely that it is the viewers, in ways thoroughly constrained by their existing linguistic, social and ideological knowledge, and usually without any overt awareness that anything might be happening, who are doing things with the media. (Stuart-Smith 2011: 223) As cautioned earlier, Stuart-Smith and her team did not conduct their research on a global scale. If they had, then there could have been a smoother transition through Approaches 1-5; that may also have lessened the significance of conventional diffusion in their findings, perhaps strengthening the empirical basis for media influence. As Macaulay noted back in 2001 (cited in Approach 1): 'how be like reached teenagers in Glasgow . . . is unlikely to have been through direct contact with young Americans' (2001: 17-18). Stuart-Smith and her team stress that the London innovations were likely already present in Glasgow, and that the media only expedited their recent surge. This prevents any extrapolation of a decisive media role for innovations that had no such prior presence, such as be like. That question remains open, and it is precisely such questions which the mediated innovation is designed to help coordinate.
Similarly, in terms of linguistic features: 'a fruitful line for further research would be to develop a more comprehensive account of the social and linguistic conditions which permit some [linguistic] elements to be more widely available than others' (Milroy 2007: 165). The mediated innovation model could be a useful backdrop for such a comparative analysis.
Lastly, to return to a caveat at the outset, the mediated innovation model does not cover the influence of the internet. Buchstaller et al. (2010) and Coppen and Foolen (2012: 271) analyse innovation use in online discussion forums, but do not specify the whereabouts of specific participants, their duration of involvement, or really much at all about them individually. That curtails any account of innovation movement across spatial or social distance. Stuart-Smith and Timmins (2009: 42) did include 'engagement with computers and the internet' as an extra-linguistic factor, but its significance was low: the only detectable result was a negative effect on TH-fronting in wordlists (Stuart-Smith et al. 2013). However, their data were collected in 2003-2004, a very long time ago in internet terms (predating the rise of social media like Facebook, Twitter and MySpace). Engagement with the internet has changed almost beyond recognition in the intervening decade, and a great deal of research remains to be done there.
My own sense is that using social media is much closer to interpersonal interaction and conventional diffusion than the communicative imbalance and parasocial interaction characterised by TV viewing. For now though, as Stuart-Smith, Timmins and Tweedie (2007: 255) put it, that is another story altogether. Jane Stuart-Smith has been very patient providing and discussing her unpublished work. She also gave useful comments on a draftas did Robert Lawson. A version of the paper was submitted for the Richard M. Hogg Prize; those reviews provided constructive criticisms. I have had further useful comments over the years from Dave Elder-Vass, Isa Buchstaller, Peter Trudgill and David Hornsby. Many helpful pointers came from conference and seminar audiences at the University of Essex (U.K.), the University of Groningen (Netherlands), the University of Cambridge (U.K.), City University (U.K.), the University of Regensburg (Germany), and the University of Kent (U.K.). Two anonymous reviewers motivated a major overhaul, including a rethink of the main neologism -'mediated innovation model' had previously been 'linguistic virtual collective', which appeared in the above conference abstracts (most of which remain online). V eronique Lacoste kindly translated the abstract. Allan Bell and Devyani Sharma have provided painstaking editorial oversight, and Trish Brothers has conducted excellent copy-editing. Remaining mistakes are of course my own.