Attitudes and Mental Model of Language : On the Cognitive Foundation of Sociolinguistic Practice

In this paper I would like to argue that we are better sociolinguists if we care about what linguistically unspoiled people think about language in other words: if we know more about people's mental models of language. Secondly, I would like to think about if and how those mental models constrain linguistic behavior. I will argue that, indeed, there is a connection between mental models and linguistic choices, but that this connection follows general principles of highlevel cognition. Since cognitive semantics is the linguistic subdiscipline which tries to find out how human conceptual systems are organized and how they are related to linguistic units, it is helpful to use ideas from cognitive semantics in order to understand the scope, effects and internal mechanisms of our mental models of language. The third problem I want to address in my paper is a methodological one: what methodological choices should sociolinguists make in order to find out more about the cognitive and social foundations of language and language change.


Introduction
In this paper I would like to argue that we are better sociolinguists if we care about what linguistically unspoiled people think about languagein other words: if we know more about people's mental models of language.Secondly, I would like to think about if and how those mental models constrain linguistic behavior.I will argue that, indeed, there is a connection between mental models and linguistic choices, but that this connection follows general principles of highlevel cognition.Since cognitive semantics is the linguistic subdiscipline which tries to find out how human conceptual systems are organized and how they are related to linguistic units, it is helpful to use ideas from cognitive semantics in order to understand the scope, effects and internal mechanisms of our mental models of language.The third problem I want to address in my paper is a methodological one: what methodological choices should sociolinguists make in order to find out more about the cognitive and social foundations of language and language change.
One of the major issues in attitude research has been the question of how attitudes are related to linguistic practice.Attitudes have been a cornerstone in social psychological research for many years now, but nevertheless there is considerable disagreement on what they are and if and how they affect human action (cf.Edwards 1994: 97;Kolde 1981: 336).I will try to use what we know from sociolinguistic and cognitive linguistic theory about how the (social) mind works in order to shed some new light on attitudes and linguistic practice.
The general issue this paper is related to is the apparent inconsistency of people s beliefs concerning languages, dialects or sociolects and how people linguistically behave in social interaction.This is an old type of question in social psychology: very early, attitude research has come up with the insight that people might expose one kind of attitude in an interview and behave differently in 'real life'.The most famous example might be the one by LaPierre in the 1930s.At a time which was characterized by overt discrimination of Blacks and Asians (not only in America), LaPierre traveled all over the US with a Chinese couple.They were being served in almost every hotel and restaurant they visited.After that, LaPierre gathered responses from the hotel and restaurant staff, asking them to state if they would admit Chinese in their establishments or not.Most of them said "no".This lead the early attitude researchers to believe that very often there is no direct link between attitudes and behavior.Another more recent study in Montréal (Bourhis 1984) has asked francophones if they reply in English when a stranger addresses them in English.They insisted much less on their native language in actual linguistic practice than in their self-evaluations.
Sometimes, this type of evidence is used as a fundamental argument against attitude studies, claiming that the results of attitude data collections do not give valid pictures of the 'real' attitudes (cf. the discussion of this issue in Vandermeeren 1996: 696).In this paper, I do not advocate this type of critique.On the contrary, I propose to accept this discrepancy as an interesting fact and I will show that it can be seen as an entailment of some basic aspects of human cognition.On the basis of those cognitive mechanisms and constraints, I will try to give a new account of the interrelated attitudes, mental models and sociolinguistic practice.

Sociolinguistic accounts of attitude and behavior
Attitudes are generally deemed to be dispositions to react favorably or unfavorably to a class of objects (Edwards 1994: 97).In a very general way, almost all sociolinguistic models of language change are implicitly assuming attitude-related causalities: for instance, speakers are likely to converge towards positively evaluated varieties.Traditionally, sociolinguistic argumentation is prestige-based (J.Milroy 1992: 149).Although some sociolinguists have been very critical about the naive use of prestige for the explanation of language change, it is still extremely common to use this concept in a rather unquestioned way in contemporary sociolinguistics.The underlying idea is that either particular social groups or some members of a social group bear more prestige than others, and that the linguistic differences between high-prestige and low-prestige people are the main motor behind linguistic change from below.The Labovian tradition postulates a 'linguistic innovator' who bears a lot of prestige due to his or her socio-economic status in the local communities.At the same time, Labov claims that this innovator has to have important ties outside the immediate social environment: Thus we have a portrait of individuals with the highest local prestige who are responsive to a somewhat broader form of prestige at the next larger level of social communication (Labov 1980: 261).
The early account of such a correlation between the attitudinal domain and language change can be found in Labovian-style correlational sociolinguistics.Although it might not be necessary, I will give a -admittedly very simplifiedaccount of the core arguments of this type of studies.Firstly, there is a continuum of prestige which is correlated with the social strata within a particular area.Secondly, there is generally some kind of a standard accent.The lower the social class, the further away it is from this standard: Fig. 1: Sociolinguistic studies in the Labovian paradigm show how higher social classes and more formal styles tend to be closer to an assumed standard (most commonly represented by the abscissa).This figure depicts the values of the seminal study by Labov on the stratification of the (r) variable in New York City (Labov 1966).CS: casual style; IS: interview style; RP: reading passage; WL: word list; MP: minimal pair list.
In a Labovian sociolinguistic framework, the consistently observable change across the different context styles mirrors the people's tendency to converge towards what is thought to be the prestigious norm.The underlying assumption is that higher social status (usually measured in socio-economic terms) gives more prestige to the variants used by higher-status people.Prestige is thus a term which denotes a value judgment closely tied to socio-economic hierarchies.It refers to at least two social entities: firstly, to higher socio-economic classes which are the bearers of prestige, and secondly, to prominent members of a particular class or group.The two are sometimes at odds with each other, e. g. a prominent leader of a street gang is not a particularly prestigious member of society on a more global level.James Milroy, in his very helpful article from 1992, discusses such different forms of prestige: On the one hand, we have a kind of prestige that is somehow agreed on by the wider community, but on the other a micro-level kind of 'local prestige', which is presumably not the same [...].It seems appropriate to inquire into what these kinds of prestige might be, as they appear to belong to two different orders of conceptualization (James Milroy 1992: 150).
In my view, Milroy's analysis points exactly in the right direction.Sometimes, if the two forms of prestige described by Milroy are at odds, sociolinguists use the term covert prestige.Trudgill's Norwich data (1974) show that there are not only social pressures to converge towards upper-class or standard varieties, but that non-standard forms can carry some sort of in-group prestige which leads to the well-observable fact that a lot of people continue using stigmatized underclass-variants.I will come back to the notion of covert prestige in section 4.3 of this paper.

Three Mental Models of Language
It is important to distinguish between the concept of language attitude as a predisposition to react in a certain way to a linguistic stimulus, on the one hand, and more conscious language-related mental representations, on the other hand.I propose to call this latter category non-linguist's 'mental models of language'.My claim is that the study of such mental models of language might shed at least some light on how attitudes arise and how we can understand the relationship between attitudes and social action.The only major research project on laypeople's ideas about language I am aware of is the one carried out by Preston and Niedzielski (1999).Other data on what people think about particular accents can be collected from various studies in the anthropological, dialectological and sociolinguistic domain, where scholars are discussing their findings with the help of field notes about their informants' metalinguistic comments.The hypotheses presented in this paper rely partly on this type of metalinguistic statements.In addition to that, a search on the World-Wide-Web for webpages and newsgroup-postings containing statements about language in general has been conducted.
One important idea I will use in my analysis is the theory of conceptual metaphor as proposed by George Lakoff, Mark Johnson and others (cf.for extensive introductions and discussions Lakoff & Johnson 1980, Lakoff 1994, Lakoff & Johnson 1999).For Lakoff, Johnson and their colleagues, metaphorical language is not a mere matter of ornamental style.On the contrary, many domains of human culture are essentially metaphorical (Lakoff/Johnson 1980: 40) and we have probably no other way of thinking and speaking about them than by using metaphors.My assumption here of course is that language is one of those domains and that metaphorical ways of talking about language can reveal how people think about language.The basic idea of the theory of conceptual metaphor is that human cognition uses ontological and epistemic aspects of an experientially accessible source domain in order to understand a more complex (and sometimes more abstract) target domain.The ubiquitous metaphor KNOWING Is SEEING for instance tells us something about how our mind transfers embodied experienceseeing -to abstract domains such as thinking (Lakoff & Johnson 1999: 48).The important part of this theory of conceptual metaphor is that the source domain does not merely offer a term for speaking about the target domain, but that it has important entailments for the way the target domain is conceptualized.Participants in the source domain can be mapped onto participants in the target domain, and depending on the metaphor chosen, the reasoning in the target domain can be shaped by the source domain.
The LOVE IS A JOURNEY mapping does not simply permit the use of travel words to speak of love.That mapping allows forms of reasoning about travel to be used in reasoning about love.It functions so as to map inferences about travel into inferences about love, enriching the concept of love and extending it to love-as-journey (Lakoff & Johnson 1999: 65).
Metaphor analysts in this framework found a considerable number of source domains linked to extremely basic experience in the sensory-motor realm.Joe Grady (1997) came up with the idea of primary metaphors as a set of mappings which we learn extremely early in our lives.Things like MORE IS UP or IMPORTANCE IS SIZE, CATEGORIES ARE CONTAINERS, KNOWING IS SEEING are mappings which are learnt through constant co-activation from one thing with another thing: a baby sees scenes where the addition of liquid or some other material leads to some kind of raising of a level or upper surface.The consistent co-activation of those two domains then leads to a conflated concept of ADDING and RAISING.The metaphor KNOWING IS SEEING is based on the crucial status of vision in early learning and in human cognition generally.The metaphor CATE-GORIES Are Cont aine rs is experientially based on the fact that things of a kind are often located in similar bounded areas or containers.This leads to the idea that the world is structured in distinct and well-defined categories.Such primary metaphors are of course a possible explanation of widespread polysemies and semantic changes such as perception verbs for mental states (e.g.I see for I understand), but, again, Grady, Lakoff and others are arguing that we are actually reasoning in those metaphorical terms.
Metonymies differ from metaphors in one basic point: we are not dealing with two different domains (source and target) but only with one domain, an entity which is internally related to the domain is taken to represent the whole domain.
mother=houswife-mother Pentagon * defense ministry Fig. 3: Metonymy as a mapping process within one single domain, where one participant or feature of the domain is used to represent the whole domain (cf.Lakoff 1994: 79f.).
The mapping thus takes places within a single domain.Metonymies are extremely important for social stereotypes, such as ideas about gender roles in many societies: from the whole domain of mothers, those who are housewife-mothers are taken as the 'best' examples and represent the prototypical mothers.Deviations from this prototype are considered to be marked and often less valuable (Lakoff 1994: 79f.).
Metaphors and metonymies play an important role in the last analytical tool to be introduced here: Idealized cognitive models (ICMs).ICMs are built up representations of a domain, and they are grounded in experience and in innate aspects of mental processing.Besides metaphorical and metonymical mappings, ICMs make use of frames and scripts, and image-schematic structure (Lakoff 1994: 68).Lakoff illustrates the ICMs with the example of the western mental model of a week: there is no 7-day cycle out there in the real world, but we have a culturally defined frame which involves the sequence of 7 days and which constitutes the ground on which a concept like Tuesday can make sense.ICMs do not only organize the noisy sociophysical world around us, they serve as normative backgrounds, shape our expectations and in some cases even determine to which differentiations we pay attention or not.Category structures and prototype effects, as described in Lakoff 1994, are consequences of the omnipresent use of ICMs.Now let's turn to the mappings where language is the target domain.It is important to note that the search was not aimed at metaphors for COMMUNICATION (as investigated by Reddy 1993), but only for those involved in the cognitive construal of the phenomenon of language itself.A list of the preliminary results of my search for metaphorical conceptions can be found in the appendix of this paper.Although many of the source domains listed in the appendix are somehow more concrete than the target domain, sometimes it seems to be possible to use very unspecific and abstract source domains as well (e.g.Language Is A System).2 2 Although many of the source domains listed in the appendix are somehow more concrete than the target domain, sometimes it seems to be possible to use very unspecific and abstract source domains as well (e.g.Language Is A System).
As I have already mentioned before, the source domain which is used to think and argue about language has important entailments for the logic which is applied for the target domain.If we think of language as a MACHINE or a BUILDING, we are imagining an internally structured artifact which can be assessed according to the quality of its makeup.If language is a machine, it can run well or break down, if it is a building, it can stand erect through earthquakes etc. or collapse.If LANGUAGE IS A TOOL, it has a high functional value for reaching particular goals in life, if it is simply home, it has a high affective component which does not necessarily coincide with its usage-value.
The target domain LANGUAGE is very hard to construct in a direct way.Conceptual metaphors function as a mechanism which allows to understand one (difficult, abstract, previously unknown) thing in terms of another.This other domain is preferably an experientially more accessible entity.Language as a phenomenon is at the same time ubiquitouswe use it constantlyand abstract -it is not something like a cup of coffee, a fjord or even the social security agency.And even linguists are having a hard time finding the appropriate expert mental models for it: we are all familiar with the competing metaphors within the domain of linguistics (just think of the idea of a language 'instinct' or 'organ').
Not all of the metonymies and metaphors listed here are equally frequent and interesting for the analyst.For our present purposes I would like to focus on a selection of mappings which I believe are central for the understanding of sociolinguistic processes.

Language Is A Physical Structure (Building)
This metaphor is probably a close relative of the primary metaphor lo gic al Org an izat ion is phy sic al Stru ctu re (ak a .theo ries are BUILDINGS; Grady 1997: 282; Lakoff & Johnson 1999: 57f.).The meta-phorical construal of language and particularly grammar as a house or other large erect building is quite common.It has a great tradition at least in Western thought, as Mittelberg (1999) shows clearly.One of the most beautiful examples is the "tower of grammar" as shown in figure : <"•?WifiT' -QltrttJrftScc ritrtjlicljcilt ffitcmMbtcjitgcliDctrerctlEitn.Even grammarians shared and were using the metaphor LANGUAGE Is A BUILDING, as the following quote by Carl Philipp Moritz supports: Allein die Sprachlehre hat einen höhern Endzweck: sie soli uns die geheimen Fugen auseinander legen, wodurch das Gebäude unsrer Sprache sich ineinander schlieBt [...] (Carl Philipp Moritz 1794: 4; emphasis RB).3 3 But the grammar book has a higher goal: it shall explain us the secret joints through which the building of our language holds together ('closes itself in').
As in many other cases, an abstract entity is understood in terms of a concrete object rooted in physical space.The physical object in the source domain shows an extremely high experiential familiarity.We all are used to move around inside of all kinds of buildings.We know how buildings can limit our range of action or our viewpoint, we have some ideas of how they rely on the laws of statics etc.There are all sorts of entailments of this type of construal, for instance a building which is limiting the speaker's motion in space, i.e. the famous "prison-house of language" (cf. the book-title by Jameson 1974).
The sociolinguistic relevance of this metaphorical mapping from buildings onto language becomes immediately clear if we turn to many people's ideas about the contrast between standard and nonstandard languages.It seems that for many informants in very different settings, officially normed languages are planfully structured buildings which are constructed in a solid way.One of the frequently encountered entailments is that you can rely on the stability of the standard whereas the non-standard is unstable and chaotic: We sort of speak a bit slang, sort of innitlike we would say 'innit' and all that.He was scared we might laugh at this perfect sort of English ... the good solid English that they teach 'em (Rampton 1995: 49;emphasis RB) In this quote from Rampton's data, an Indian adolescent is talking about his newly immigrated cousin who had benefited from upperclass education back in India, and whose idiolect is much more upper-class than the one of the local Indian adolescents in the area.
What we find here is thus the idea that 'good' standard language is a solid construction, as opposed to the unstable non-standard.I came across this kind of idea when doing fieldwork in Southern Germany: many of my informants (mostly farmers) were flabbergasted that the fieldworker insisted on the fact that their home-variety was worth studying and even was supposed to have a consistent grammar.However, we can also find similar data in the very different setting of the Mayan language Jacaltec (spoken mostly in southern Guatemala; cf.Grinevald Craig 1979: 52): So the Jacaltec speakers also were very curious and puzzled about the intrusion of a foreign linguist who presumed to tell them that their language indeed has a grammar, as do all languages, and that it was well worth studying.Some were apologetic, saying their language had broken down, accusing themselves of not learning it and respecting it as had their parents and ancestors.
The Jacaltec speakers thought that only Spanish had a grammar.There are observations of both the insider view and the outsider view about dialects having no grammar.Finally, Niedzielski & Preston (1999: 22) encountered the same phenomenon in the North-American context: The abstraction is rule-governed; the deviations are not.The linguist's so-called rules of AAVE or lower-class New York City English are, therefore, nonsense.What rules could there be when the forms under discussion are simply failures to observe the rules of "The Language"?In more linguistically familiar terms, such varieties for the folk are performance deviations from competence, not alternative competencies (Niedzielski & Preston 1999: 22).
In terms of the metaphorical model I am proposing here, the standard language is construed as a stable physical structure, whereas dialects are an unstructured chaos of sounds and words.The normativity, attitudes and sociolinguistic processes which are entailed by this cognitive mapping are well-known to many sociolinguists.The source domain BUILDING allows vivid mental imagery in its elaborated metaphorical use: in a very colorful way, people can express their attitudes towards particular languages and dialects.Buildings can constrain our movements, can give us shelter, can collapse suddenly or withstand the threats of a hostile environment.All those ontological and epistemic details of the source domain bui ldin g can be mapped onto language: it can be a communicative barrier, it can be a threat for (minority) cultures or protect them.Of course it does not always have to be the buildingsource domain, it can be any other structured physical object or even the quite abstract source domain of a system.However, all those mappings have the same type of entailments in the target domain of language.
However, not all societies are characterized by such anti-dialect ideologies.The next metaphorical mapping is among other things relevant for attitudes towards dialects in dialect-friendly places like Norway.

Language Is A Raw Material / Natural Resource
This is an important metaphorical mapping which affects directly the way people evaluate linguistic varieties.It is the driving force behind all forms of linguistic purism.The source domain Raw MATERIAL or NATURAL RESOURCE has crucial inferences for the target domain LANGUAGE: a language has a high value, if it is pure.Purity is the initial state of language, and through the negligent use of the people it becomes more and more unpure, mixed, alienated.Contaminated, mixed, unpure materials are less valuable in most manufacturing processes, and purification and refinement are expensive, energy-consuming and sometimes even technically impossible.If we think of language in terms of this source domain, we are making strong inferences about such purity-values which are in direct contradiction to everything linguists should know about language: mixing and contact are not phenomena of secondary importance but rather central aspects of any natural language in any place of the world and at any time.Even linguists sometimes tend to forget that not the contact-induced phenomena of change are the exception, but rather stasis (cf. the debate between Muller and Schuchardt as rendered in Thomason & Kaufmann 1988: 5f.).BOUNDED SPATIAL REGIONS (Grady 1997: 283).This, for the case of languages, means, that a language has to be clearly in one or the other container and cannot be in both.A real, good language has to be pure, mixing is bad.Again, there are prototype effects at work (Lakoff 1994: 68).One prototype of the category language is the ideal language: a 'pure' idiom which shows an ideal state of systematicity, is not affected by decay due to bad usage, shows no mixing etc.The widespread belief that language is in constant decay due to mixing, careless use, and other external influences resides on this Raw MATERIAL metaphor.The epistemic entailments of this mapping are that language used to be "pure" and "good" in earlier times and maybe still is pure (in the case of dialects) in remote, isolated communities (cf.Berthele 2001a).The ad depicted in fig.above uses this metaphorical mapping in order to create a negative attitude towards multilingualism.Other examples can be found in accounts from anthropological linguists (Woolard 1998: 17), or language acquisition studies, where bilingual children as young as 2 show selfcorrecting when they are mixing languages (Foster 1990: 195).The very popular (and in my view quite unnatural) idea of 'one personone language' (cf.e.g.De Houwer 1995: 246) as a principle for bilingual parenting has roots in the very same cognitive inference.
The Language is A na tu ra l Res ourc e mapping interacts with the principle of granularity (cf.section 4.1 below): depending on the level of granularity a particular setting activates, a standard language or a local dialect can be seen as 'pure'.A while ago, I was talking to a Spanish cognitive linguist who was extremely surprised to hear that there are places where people think of DIALECTS as being pure.For her, only standard languages were candidates for linguistic pureness.Dialects are per definitionem impure.

Language Is A Territory
I think that this mental model is not a metaphorical mapping from one domain onto another but rather a metonymical process: one entity out of a framing domain is taken to represent the whole domain.Since languages typically are tied to groups of speakers living in certain geographic areas, those languages can be construed as representing particular areas.Additionally, certain aspects of a language can be seen as iconically reflecting the topography of the land where its speakers are living (cf. the concept of iconicity in Gal & Irvine 1995: 973f.).Languages are often construed as 'sticking' to particular places, as the following example from a newsgroup posting shows clearly: J'ai appris il y a longtemps, qu'il y avait plus d'Anglophones que de Francophones å Montreal.Je trouve fa effrayant de voir une ville Québécoise, qui ne sait pas maintenir une language å sa place!! Que personne me dise que Montréal est une ville qui refléte la langue Francaise, ce n'est pas vrai.4 4 Source: http://www.iagora.com/I learned a long time ago that there are more Anglophones than Francophones in Montreal.I find it horrible, to see a city in Quebec which cannot maintain a language at its place!! Nobody tell me that Montreal is a city which reflects the French language, that is not true.
This posting reveals nicely the idea that languages belong to places, and that it is the people's duty to maintain them where they belong.
Even if nobody can reasonably deny that languages stick to people rather than to places, this metonymical mapping has a strong effect on people's ideas about languages.
The mapping of a language onto a territory has to be seen as the core of a very powerful Idealized Cognitive Model (ICM) of territoriality, as described by Rubba (1996: 241): This ICM must include some sort of definition of a culture as a group of people related by ethnicity, beliefs, language, cultural practices, etc. [...] Most directly relevant is the idea that members of the same culture live in spatial proximity to one another, and that the space they occupy is contiguous.The space they occupy is then their territory, i.e., their physical space is conceived of as belonging to them.No more than one culture can occupy a given space in the model (Rubba 1996: 241).
This model thus involves metonymic relationships between language, culture, territory, and social network.This ICM of Territoriality is closely related to the romantic or Herderian conception which equates language, ethnicity and nation (cf.Coulmas 1988, Dorian 1998).Among many nations which somehow represent this ICM, France is an excellent example where the national ideologyat least since the French Revolutionis strongly focussed on a combination of national territory and language as a national heritage and resource.All non-standard variants of Romance languages and all non-Romance languages are long dead or have an extremely marginal status.But even in much less prototypical nation-states, the same ICM applies.In multilingual Switzerland, e.g., there is an official political principle of territoriality (cf.Rossinelli 1992: 179, Berthele 2001b) which means that at least three of the four national languages are intrinsically tied to particular areas on a sub-national level.This leads to the characteristically underdeveloped status of bilingual institutions (schools, media, etc.) in a country generally seen as genuinely multilingual.The German-, French-, and Italianspeaking areas are then seen as monolingual, i.e. schools are in local language, even if there are important minorities of other national languages.Due to this ICM of territoriality, this coexistence can last for hundreds of years without any strive for setting up bilingual education.5This is why I think Switzerland shouldparadoxicallybe labeled as a genuinely monolingual country with four national languages.There are competing criteria to attribute a particular territory to a particular language, sometimes it is the demographic reality, sometimes its historical affiliation.It is not surprising at all that such a linguistic territorialism goes together with the extreme and often quite dysfunctional form of federalization in Switzerland (e.g. 26 different educational systems).6 5 Speakers of the smallest language, Romansh, are all bilinguals with German and the model is thus slightly different for this area.6 In order to maintain a minimal internal multilingualism, the national policy is to teach one of the national languages as the first second language in school.With the rising importance of global English, there is a general desire to teach English as the first second language, which has provoked a big debate in recent years (cf.Mittler 1998): particularly the Francophone minority fears that the German-speaking majority might be turning its back completely to French and shift totally towards English as the most important foreign language.
Even for genuinely multicultural places like the US, the ICM of territoriality is crucial.The leaders of the English-Only-movement I have been talking about before are arguing in a very territorial way: The US-English ideologues can of course not assert that there is only one culture living in the US.The cultural differences of the different immigrants and natives are acknowledged and in the same time erased by the evocation of "the American Dream of economic and social advancement" (English Language in Public Schools 1998) or the notion of the famous "melting pot".So, for the campaign in favor of Official English, the ICM postulates one nation, admittedly multicultural, but bound together by common values such as the American dream, probably freedom of speech etc., and one language.
What is it that has made a society out of the hodge-podge of nationalities, races and colors represented in the immigrant hordes that people our nation?It is language, of course, that has made communication among all these elements possible.It is with a common language that we have dissolved distrust and fear.It is with language that we have drawn up the understandings and agreements and social contracts that make a society possible (Hayakawa 1994: 15).
Again this seems to be a prototype effect based on the idea of an 'ideal nation': One ethnic community living in one coherent territory, with one common language.The additional metaphorical mapping here is that society is seen as a bunch of objects which have to be held together by something.This something is the common language.The ideal nation, just like the ideal husband, does not exist in reality, but its cognitively central status has important consequences for normative judgements about actually existing nations out there in the world.
So why not imagine English as a national means of communication?Again, the principle of territoriality seems to inhibit this solution: since English is an external language without a seat on the national territory, many Swiss find it odd to use it in the national context.

Mental Models and Actions: Granularity and Attitudes
In the preceding sections, three of the most important metaphorical and metonymical mappings involving the target domain LANGUAGE have been presented: LANGUAGE AS A PHYSICAL STRUCTURE, as a RAW MATERIAL, and as a TERRITORY.In this section, the consequences those mappings have for sociolinguists will be spelled out.Thus, the main question here will be how those mental models of language affect language usage.
There is no doubt that the ICM of territoriality affects the way people evaluate languages and dialects.Given the common idea that a particular language belongs to a particular area, it is not very daring to expect other languages to be considered marked choices in that area.The metonymy LANGUAGE Is A TERRITORY undoubtedly plays an active role in the background of both language attitudes and linguistic behavior.But this ICM of territoriality is a moving target.In section 2 I have argued that there seem to be different types of prestige which often are at odds with each other: there is a tendency to chose one variety as a global prestige-variety, associated with socio-economic class.And there is a tendency to see prominent members of the peer group as prestigious speakers.This problem can be better understood when considering what I propose to call the principle of granularity.

Principle of Granularity
Both forms of prestige J. Milroy (1992: 150, cf.quote in section 2) are instances of prototype effects within an ICM of territoriality.I propose to analyze what Milroy quite vaguely calls "different orders of conceptualization" in terms of different levels of granularity.Granularity is the degree of resolution which is applied to a particular area of perception or conception: The parameter of granularity applies relative to a particular level of scope.Granularity is the coarseness or fineness of the grid with which one attends to the contents within the chosen scope.That is, it is the general relative magnitude of the subdivisions that result from the further partitioning of the chosen scope of material (Talmy 2000: 456).
The following example should illustrate the application of different levels of granularity in discourse.It is passage from an interview that has been conducted with a nine year old informant (I) in the bilingual town of Fribourg (cf.The topic of the interview at this point is the linguistic and social environment the child is situated in.This passage shows very clearly how both interviewer and interviewee start at a very low level of granularitythe default level in this particular setting is about if somebody belongs to the French-speaking majority or to the German-speaking minority.The label "tutschi" which actually means Germans is used.This is a case of metonymy, since the informant does not mean Germans from Germany, but Germanspeaking people.Subsequently, we both zoom in to the level of "tiitschwizer", Swiss Germans, i.e. the types of German are more and more specified (German -Swiss German -Fribourg German).In many interviews I have conducted, the level of granularity can be even higher, namely at an idiolectal level, when informants characterize their idiolects as strange kind of Berne German or mixed Fribourg German.
Figure is supposed to depict in a schematic way the range of different vantage points speakers can take when it comes to construe the "chosen scope" (Talmy 2000: 48) of language.Note that the different levels of granularity are all representing different variants of the ICM of territoriality.Let's start at the most general level of this figure.In many contexts, an idealized prescriptive norm is figuring on the lowest level of granularity in the British context, e.g., something like RP, standard French in France, and Standard High German in Germany.7In Labovian sociolinguistics, this level of granularity might correspond to the assumed, most 'general' prestige-variety (cf.section 2).
A speaker who is assessing linguistic variants in terms of prestige might zoom in a little more and stop at a higher level of granularity, maybe at the level of the region.Thisadmittedly unspecificlabel indicates that now it is not the national level but rather some regional entity which represents the scope.In Germany, this might correspond to the level of varieties such as Schwäbisch (Suabian) or Bayerisch (Bavarian).For places with widespread use of traditional dialects, this level corresponds to the basic level of categorization (Christen 1998: 261).The basic level categories represent varieties at a substandard-level which are associated with a gestalt-like mental image, they are seen as internally homogenous categories.In German-speaking Switzerland, the basic level of meta-linguistic categorization corresponds to the Kantonsmundarten, i.e. the dialects associated with corresponding cantons (e.g.Bemdeutsch, Zurichdeutsch, Baseldeutsch, etc.).
Note thatfor the principle of granularitythe notion of physical space should not be taken too literally.Especially on higher levels of granularity the relevant 'space' might as well be 'social space', such as a speaker's primary dense and multiplex network.There is no doubt that people construe their social environment in terms of closeness and remoteness, and it is a truism that sociopsychological closeness usually goes together with closeness in physical space: we do not like to spend too much time with people we dislike profoundly, on the contrary, we want to be physically close to people we are socially close.
The limiting case of application of the ICM of territoriality is the dyad, a two-participant interaction.This is the maximal level of granularity, and the territory associated with two single speakers is very small.
On every level of granularity in such a continuum we can observe prototype effects: some groups or individuals are seen as good examples of the respective variety, others are seen as representing it in a less prototypical manner.In traditional dialectology, the oldest and most un-mobile informants represented the protoypical speaker, a speaker expected to mirror the 'real' variety of a particular rural community (a variety the German dialectologists call Ortsmundarf).In Labov's sociolinguistics, the influential personality within a peer group is another example of such a prototypical speaker, in this case on an even higher level of granularity.
The significance of particular levels of granularity might vary across different places in the world.Which level of granularity is foregrounded for an individual or a group of individuals depends also on situational constraints: we can observe that institutional contexts as encountered in schools, universities and other official instances often seem to trigger a construal on a lower-level of granularity -standard languages are often perceived as more appropriate in those situations.On the other end of the scale, familyand close friendship contexts favor a construal of high granularity.Maybe, in Norway and German-speaking Switzerland, relatively high levels of granularity are more foregrounded in the mental construal of linguistic diversity than in other places.But at any level of granularity, there are always the same mechanisms of construal at work: 1. Prototypical speakers represent prestigious way of life and use the prestigious language 2. In the case of a positive identity-relationship between ego and the prototypical speaker of a given level of granularity: a) common features are highlighted, differences are erased b) varieties close to prototype evoke positive attitudes 3.In the case of a negative identity-relationship between ego and the prototype of a given level of granularity: a) common features are erased, differences highlighted b) varieties further away are linked to positive attitudes 4. The community represented by the prototypical speaker is construed as culturally/linguistically homogenous In every sociolinguistic setting, there is a default level of granularity.This is the level on which we tend to see common ground in the construed in-group and differences with respect to the out-group.
The level which is active by default in a particular setting is not by definition the basic level of categorization, but nevertheless we expect basic level categories to adopt a salient position within the mental construal of linguistic diversity.In the discourse example from Fribourg given at the beginning of this section, the default level is on a lower level of granularity, i.e. on the level of German vs.
French.The primary distinction in this town seems to be if a given person is either German-speaking or French-speaking.But the same discourse example shows also that the zooming in from this very low level of granularity is very easily done, even by children.
In terms of Milroy's network model (cf.L. Milroy 1992), we can assume that in settings with high network density and multiplexity, relatively high levels of granularity play a profiled role: in-group norms are dominant and violations of those norms extremely marked.On the other hand, in settings where weak ties are dominant, traditional local variants can become the marked choices, even if we can assume that speakers genuinely have very strong positive feelings with regard to their native in-group dialect.Maybe it is simply the functional need to communicate outside the immediate in-group network which triggers a different construal, i.e. in this case a lower level of granularity.Maybe the frequent interaction with out-group-people also has some affect on the attitude an individual might have towards his or her own primary network.Prestige attributions eventually might follow this new need for a larger communicative scope: the contact with out-group-people was motivated by functional need, and it entails new solidarities outside the former dominant in-group network.In any case, speakers can adapt their construal of granularity very easily, depending on the type of situation their thinking is referring to.
The different components of the folk model of language tend to affect different levels of granularity in different ways.If the 'solidity' of the language-building is at stake, people are usually thinking on a quite low level of granularity: the logical makeup is an issue important for standard languages.It is the high-prestige people in science, culture and politics who tend to use standard varieties in both the written and the spoken mode.But in some countries, people are discussing anything in dialect, including quantum physics and generative grammar.This is the case in Germanspeaking Switzerland, although only in informal situations as opposed to lectures at the university.It is also the case in Norway, where many people are using dialect in virtually all contexts.These are relatively unusual situations in the western world, and we can observe several types of critique of this distribution of use of dialect and standard.
Firstly, some people claim that dialects do not provide enough subtle linguistic tools in order to talk about complex matters.Even though this argument is proved to be false by the simple fact that people are able to talk about anything in dialect effortlessly, the idea of dialect as a shaky hut, a structure without rules and logic seems to be stronger than any real-life evidence.Thus, we observe once more the application of the mental model LANGUAGE Is PHYSICAL STRUCTURE.
Secondly, a related argument against the use of dialect particularly in educational contexts relies on the Langua ge Is A Nat ura l RESOURCE model: The fear that too much dialect-use in formal contexts hinders the production of good standard-language in writing.Scholarly texts which show features of regional language varieties do not correspond to the level of formality 'pure science' has to represent: Pure scientific prose should not have the smell of rural muckheaps.
Thirdly, some people fear that if we start using dialects in scientific and administrative contexts, those varieties are bound to lose their authenticity and start converging to some kind of an 'unpure' pseudo-koiné.Again, we encounter thus the LANGUAGE IS A Natu ral Res our ce model.
The second position can be accompanied by the first position, but it is not a priori necessary: it is possible to be aware of a dialect as at least as complex a building as the standard and nevertheless wanting to separate the two worlds.Maybe just for the sake of the possible negative audience reactions to non-standardness in scientific texts.The third position finally is the one of the dialect-protectors who are arguing at a relatively high level of granularity.They are applying the LANGUAGE Is A NATURAL RESOURCE metaphor on the level of dialectsand maybe on the level of the standard, too.

A Case Study: Granularity and Prestige in the Class Network
In this section, I propose to analyze some sociolinguistic data in terms of different levels of granularity.The data stem from the same study as the interview quoted in section 4.1.The sociolinguistic setting is a protestant private school in the traditionally catholic bilingual town of Fribourg (Berthele 2000a(Berthele , b, 2002)).To be very brief about the sociolinguistic setting, we can say that the strong language on the local level of granularity is French, the strongest minority language a local dialect (Sense or Freiburg German) traditionally used by the local German-speaking minority.The town has a strong catholic tradition, that is the reason for the existence of a protestant private school: the immigrated protestants didn't want to send their children to the public school dominated by the catholic heritage.On the level of granularity of this private school there is a particular dialect which is the prestige-variety, namely a dialect which resembles the neighboring Berne area dialect.There are cultural and sociological reasons for the choice of this prestige-variety: Berne is the place from which most of the protestant immigrants came during the 19th and 20th century.Although the Bernese are not a majority in this school anymore, the Bernese dialect still seems to be the school-internal prestige variety.In my research I analyzed the dialectal variation within a class of 14 schoolchildren.Figure depicts the sociographic structure of the class as well as the values for one particular dialectological variable.The social structure within the class has been elicited using a sociogram (Moreno 1954: 34).This sociogram has been constructed based on the question "who do you like to play with".All mutual choices are represented by straight lines.The figure shows clearly how, on the one hand, most of the children share two or more lines and, on the other hand, five of them show only one or even none of those mutual sociographic choices.
Since the children in this private school do not come from traditional local families, they speak all kinds of non-local dialects and languages at home.Most of them are at least bilinguals or bidialectals, some are tri-and quadrilinguals.The linguistic variation within the class can be assessed in terms of how 'Bernese' a child speaks.Out of the 18 dialectological variables I have analyzed, figure only presents the one of the vocalization of /!/ (as in Bernese [bau], most other Swiss varieties: [bal]).This is one of the very salient features which distinguishes the Berne dialect from many other Swiss German dialects.The higher the values for this variable, the more Berne-dialect variants are found in the children's realizations of /l/.
The figure shows clearly that higher values for this variable correlate highly with the degree of integration into the class structure.On the whole, for all analyzed dialectological variables, there is a significant tendency to using Berne dialect variants with increasing social integration into the class network.But not only that, as I have shown elsewhere (Berthele 2002), there is even a tendency to lose Bernese variants which were 'inherited' from the child's family background if the child does not identify herself with the mainstream part of the group.The data even suggest that the language-biography of a child is actually a very unreliable predictor for the child's actual linguistic practice in class, whereas the variables from the sociogram-domain turn out to be excellent predictors for the degree of 'Berneseness'.This is of course exactly what the sociolinguist would expect.But the point here lies not in the mere correlation of sociographic and dialectological variables.What fig.
shows us is that the notion of a prestige dialect is relative to a particular level of granularity.On the level of the class as a whole, Bernese seems to be the prototypically chosen language by the majority.But if we zoom in to a higher level of granularity, for example the one of the friendship-dyad Yves and Martin, we can assume that Bernese here is no longer the prestige variety.This dyad, just like the one between Judith and Jessica, is much less oriented towards the class-language.In one of the interviews I have conducted with him, Martin tells me that for him the class is too small and that he does not find enough interesting friends within this group of children.Consequently, for Martin there are important additional ties outside the school-network.This latter point seems to be important for the very isolated Benjamin, too.
In terms of my model of granularity this means that we have to zoom in to the maximal possible granularity in order to see which variants and varieties evoke positive or negative attitudes and reactions.The friendship-dyad Martin-Yves seems to stabilize itself on a certain distance from the school-prestige dialect.Something similar happens in the dyad Judith-Jessica, here even with occasional switching to Standard High German (i.e.Judith's native variety of German and the only German Jessica speaks).What appears to be marked choices on a lower level of granularity can be a perfectly unmarked pattern of variation within this dyad.This is a good example for the extreme situation-relatedness of the markedunmarked distinction, an example for the multidirectional character of prestige-orientation. Linguistic prestige emerges in social interaction and is tied to preferred interaction-partners. Due to similar maximes of social action for a larger group within the class network there is a certain convergence towards a class prestige variety.If we only look at this group of children through the lens of correlational analysis, we find the expected high correlation of integration and linguistic conformity.Only if we focus on a high level of granularity on the interesting individuals, we discover the additional revealing details about competing prestige of sub-groups, dyads, and network-external actors.And only if the sociolinguists combine the analysis of high-and low-level granularity, they are able to make realistic guesses about the mechanisms behind language change on the mass level, i.e. on the lower levels of granularity.

Principle of Granularity and Linguistic Choices
At the beginning of my paper I have indicated the fundamental problems one encounters when comparing results of attitude-studies and actual social behavior.My claim here is that we can easily account for this difficulty in terms of the model of granularity: when Québeqois are asked if they reply in English when approached by an anglophone stranger, they are construing a situation on a low level of granularity, with all the ideologically constrained negative stereotypes about anglophones, the fear of Québec being anglicized, etc.When observing how Québequois actually behave in real encounters with anglophones, we are dealing with the limiting case of maximal granularity: the dyad.This does not mean at all that the ideological concepts from higher levels of granularity are gone, but they are cognitively backgrounded.What counts is the immediate situation, the ad-hoc contact to a stranger with all the insecurities of undefined role relationships and possible interactional gains or losses.
Normative attributions of prestige and stigma as in correlational sociolinguistics are only one part of the attitudinal ground on which (socio)linguistic action takes place.For the individual linguistic choice a particular speaker in a particular situation has to make, there are a lot of other influencing factors.On this low level of granularity, the refusal to speak English can have risks which are too high to be taken, e.g. the complete breakdown of communication.And it is useful to assume that the default setting of communication involves a basic will to co-operate (cf. Grice 1974).Thus, the extremely marked choice of a Québecois from Montréal refusing to speak English is rarely made.Since, in this setting, we are definitely not dealing with a in-group type of situation, it is much more likely for a speaker to accommodate to a certain degree to the other participant, maybe in something Myers-Scotton (1980: 361) calls an exploratory choice.This is why Bourhis (1984) comes to those apparently contradicting results: the attitude is global, the actual linguistic negotiation is extremely local.The analyst can only understand the actual driving forces behind the linguistic choices actually taken if he or she zooms in on the highest level of granularity.
It should be clear by now how this idea can be applied to all kinds of other sociolinguistic results: Trudgill (1983: 177) found after a series of self-evaluation tests that the norm at which a large number of Norwich males are aiming is nonstandard WC speech.This favorable attitude is never overtly expressed (Trudgill 1983: 177).
Trudgill calls this positive attitude with regards to WC (working class) speech covert prestige.Although this term has had considerable success in the international community, covert prestige is quite a problematic notionbecause there is probably nothing covert about it, or, as I will argue here, because its covertness is a matter of level of granularity.
In Trudgill's study, the term covert prestige seems to make sense because, in the Anglo-Saxon context, the analyst has his vantage point fixed on an arbitrarily (or better: ideologically) chosen level of granularity, in this case on the level of standard British English.The following quote from a very popular sociolinguistics textbook confirms the observation that there seems to be a tacit transatlantic agreement on the default level of granularity: The forces favoring the standard are crystal clear: middle-class parents talk about "good" language, school teachers correct the usage of students, letters to the editor deplore slips away from prescribed usage (Chambers 1995: 221).
Of course, in some contexts there might be good reasons to take this particular stance.But there is a certain danger that this point of view is maintained despite a lot of evidence for its insufficient explanatory power.The point is that covertness and overtness are phenomena which are constrained by the levels of granularity set up by three instances: firstly the general sociolingustic setting, secondly the immediate sociolinguistic interactants, and thirdly by the fieldworker.
Let us consider again a non-Anglo-Saxon setting in order to make the point very clear: In German-speaking Switzerland, it is not seldom to find overt normativity with regards to non-standard languages.When two dialectologists in our department were working on a dictionary of a Swiss German dialect, the local dialect protectors were furious that those 'people from the university' were including the variants of the younger generation as well: the young are not seen as prototypical speakers of the dialect, and the mere idea of including linguistic variation in a dictionary was considered absurd.You can find extremely overt normative statements about non-standard if you are eliciting them on the right level of granularity.Thus, if we return to Trudgill's Norwich study, can we really say that the Norwich men really never overtly express their attitude towards WC variants?-The fact that they do not do it with regards to the academic researcher only shows that they are aware of the normative level of granularity this type of person brings in.
In his sociolinguistics textbook, Chambers states that the pressures that maintain the non-standard "have no identifiable lobbyists" (Chambers 1995: 222).Again, I think that this is not true for other contexts.We could certainly use Norway with its strong pro-dialect and pro-variation ideology as a counterexample (cf.Romaine 1997: 10), but for the present purposes I will stick to the situation I know best, the one in German-speaking Switzerland.Here, people certainly use dialects, but they also use their variant of the German standard language ("Schweizerhochdeutsch").If the Swiss media anchors start using too much of a Germany-style accent in their Standard German, there are a lot of negative reactions particularly from academics, i.e. people of clearly upper-middle class (Löffler 1991: 44).An attitude study by Hove (2000: 156ff.) in German-speaking Switzerland has shown, that the Swiss variant of the Standard Language scores high for 'sympathy' whereas the Standard German accent scores high as the 'good' language (157f.).This corresponds to what we traditionally call the cognitive and affective components of attitudes: on the cognitive level, people attribute a higher value to the German-style accent, but on the affective level, it is the local variant which wins.
But even within the Anglo-Saxon area, we find overtly positive evaluations of non-standard speech.Consider this meta-linguistic quote from Ben Rampton's data, where an adolescent is talking about his peer-group's creole-influenced language: well, we you know think quick, we got... they you know ... only some of the girls, some of the girls are ... like the posh ones they know what to say innit... (but) some of them don't know the future language you see, we do ... they only know the past, the're they history you see (Rampton 1995: 57).
Thus, I conclude that the notion of covert prestige represents only one possible configuration of the interplay of different levels of granularity, admittedly a very typical one in the sociolinguistic contexts the leading sociolinguists are working and living in, i.e. the Anglo-Saxon one.But I think that we need a better theoretical understanding of how different forms of prestige are related.It is necessary to find a more general account of attitudes, prestige variants and speaker reactions, an account which takes into consideration the way people are construing linguistic settings and corresponding prototypical styles.8The attitudinally relevant level of granularity in the case of the men in Trudgill's Norwich sample is higher than the one taken by Trudgill and others, it seems to have a focus on more local varieties than the roofing standard.8 In the German dialect-standard continuum, e.g., we observe very often convergence towards the next lower level of granularity (i.e. the next higher level of generality) and NOT directly towards the standard language, which should 'officially be the bearer of prestige.
We can formulate these conclusions in terms of the metaphorical mappings and levels of granularity I am proposing in this paper: the Standard language as an abstraction on the lowest level of granularity may well be the bearer of 'overt' or 'official' prestige, this is the often institutionalized instantiation of LANGUAGE IS A PHYSICAL STRUCTURE.On the more local level, a higher granularity, language has to have an affective component which allows people to 'feel at home', that's a LANGUAGE Is HOME variant of the ICM of territoriality.
Thus, as opposed to the standard sociolinguistic theory, we have to be aware that covertness is not a necessary feature of the prestige of the non-standard.And, after all, in order to explain linguistic convergence or divergence phenomena, this type of expressed or unexpressed prestige can only serve as an explanation if it is active on the high level of granularity of interpersonal interaction.
The crucial issue for us sociolinguists is that actual linguistic action takes place almost exclusively at this highest level of granularity.It is in direct face-to-face interaction we are constantly shaping our linguistic means of expression according to a shared set of maximes of communication.This is the reason why global attitudes as elicited by standard attitude research are only of limited use to the sociolinguist who would like to explain language change as related to social prestige factors.
I want to close this section by stressing the importance of what Rudi Keller (1994) calls methodological individualism: Given the fact that our mental models of language arejust like any other folk models (Merz 1982)inconsistent and genuinely of a 'bricolage' type, given the fact that there can be considerable differences between low-level-granularity categorizations and high-levelgranularity social interaction, it is extremely important for the linguist not to confound those different levels.If we want to try to explain language change as something whichin addition to language-internal factorshas to do with social prestige and attitudes, and I think we should try to do this, we have to include this very highest level of granularity.We have to understand how the individual speaker-hearer with a particular linguistic repertoire in a particular setting with one or a group of other speaker-hearers makes linguistic choices.The 1000$-question then is how we comein reality and in linguistic modeling -from this intra-individual level to the collective level ofsay -the great vowel shift in English.

Conclusions
To sum sum up, there are three major points this paper tries to make.Firstly, traditional attitude-studies1 results are depending on the level of granularity activated in elicitation situation.Their value for the analysis of actual linguistic practice is thus to be relativized: if the elicitation situation does not match the real-life granularity, the results have only an indirect connection to the linguistic choices made in social interaction.Although the ideologies which are active on a lower level of granularity are present in the high-level situation of social interaction, they tend to be backgrounded in favor of more immediate evaluations of possible interactional losses and gains.The Gricean principle of co-operation and other maximes of communication often simply override the choices which would be expected on the basis of the low-level ideologies.Attitude studies often lack this congruence of granularity for data-collection on the one hand and 'action-measurement' on the other.The elicited attitudes were probably real, but not dominant in the setting of the actual social performance.E.g. in general, Americans in the 30s didn't want to serve Chinese people, but in face-to-face interaction they tended to see the particular Chinese in front of the hotel counter as almost normal people.
Secondly, depending on level of granularity and situational context, different linguistic forms are evaluated differently: this is what we might also call relativity of prestige.Different parts of the folk model of language are being applied on different levels of granularity.On the lowest level, aspects such as the stable construction of the standard language are being mapped.On higher levels of granularity, mappings such as language as a part of cultural and territorial identity are more profiled aspects of the mental model.Mental models of language can thus be inconsistent concerning what is good-bad and what is sympathetic language or not.They are no exception to the rule that folk models are genuinely instances of cognitive 'bricolage' with a great deal of inconsistencies and the very local and metaphorical logic demonstrated in section 3 above.Undoubtedly, attitudes on very low levels of granularity do exist, but when it comes to actual behavior in real-life situations, many aspects of cognition are very local, and tend to change depending on the situational give and take.Therefore, if in a given country we encounter in general very positive attitudes towards local dialects, this does not mean that there is no convergence towards the standard or towards some other kind of a regional koiné: if, due to increased geographical mobility, more and more contacts take place with out-group-members (weak ties), those contacts, with their highlevel-granularity constraints I have outlined in this paper, can lead to linguistic accommodation phenomena which go against the expected dialect maintenance.In other words, despite the overtly expressed positive attitude with regards to local norms, the actual practice leads to phenomena like dialect leveling and koinéization.I believe that by using the principle of granularity outlined in this paper, we can better understand the observation that ideologies and attitudes on the one hand do not match actual practice on the other hand.
Thirdly, the often indirect relation between attitudinal factors and linguistic practice, have some important entailments for sociolinguistic research.Even the strongest ideological and attitudinal dispositions have to be relativized by considering the maximes of communication active at the highest level of granularity.In order to explain collective phenomena of language change, the researcher has to zoom in in order to see how small changes in linguistic practicein their cumulation -affect a dialect or a language as a whole.The principle of granularity applies not only to the construal of language and social space of the speakers, but it is also a tool which allows the researcher to be aware of the level he/she is actually focussing on.Since the cognitive effort to jump across low and high levels of granularity is very low (cf. the example in section 4.1), we tend to forget that there is only an indirect connection between low-level ideologies and high-level linguistic interaction.Methodological individualism as proposed by Keller (1994) appears to be a good way to access the constraints on the most action-relevant highest levels of granularity.
Fig. 2: Conceptual Metaphor as a mapping from a source domain onto a target domain (cf.Lakoff 1994, Lakoff & Johnson 1999).
Fig. shows a particularly telling example of this inferential reasoning:

Fig. 6 :
Fig. 6: Different levels of granularity for the mental construal of language.