A Critical Linguistic analysis of the representation of Muslims in The New York Times

The aim of this study is to shed light on recurring lexical and syntactic features that contribute to a stereotyped image of out-groups in newspapers. The focus of the study is on articles relating to Muslims in The New York Times. The analysis is based on the analytic paradigm of Critical Linguistics (CL) and Corpus Semantics (CS). The results show that the linguistic features analysed point to a systematic ‘othering’ and stereotyping of Muslims as compared to other participants. The study concludes with a discussion on how the grammatical features examined work together to project a stereotyped image of Muslims and how the analytical method of Critical Linguistics (CL) copes with a quantitative analysis of a great deal of randomly chosen data from a corpus consisting of newspapers from the New York Times.


Introduction
The discourse that the media produces often functions as the foundation on which people base their conceptualisation of the world. In democratic societies the media has a very prominent position in that it checks those in power and consequently becomes a powerful actor in society. Another reason for its prominent position is that the media is considered as independent and objective. Accordingly, it often functions as the only source of information to people, especially as regards news about other societies.
After 9/11 the political and media focus on Muslims has been quite extensive. The negative representation of Muslims is often associated with the 9/11 attack. Although people are aware that Muslims have been perceived as an out-group in the West also before 9/11, the stereotyped representation of Muslims in various discourses is to a large ex-tent associated with 9/11 and thus obtains some sort of legitimacy because of an underlying assumption that Muslims, as it were, brought it on themselves. According to Said (1978) the Orient and Islam have in the West functioned as the constant 'other' "by setting itself against the Orient as a sort of surrogate and even underground self" (p 3) …"which puts the westerner in a whole series of possible relationships with the Orient without ever losing the upper hand " (1978:7). This ideological state of affairs is not recent or due to 9/11; instead as Said (1981) expresses it; the media coverage of Islam and Muslims "has given consumers of news the sense that they have understood Islam without at the same time intimating them that a great deal in this energetic coverage is based on far from objective material. In many instances "Islam" has licensed not only patent inaccuracy but also expressions of unrestrained ethnocentrism, cultural and even racial hatred, deep yet paradoxically free-fl oating hostility"(xi).
The media is a major source of information for ordinary people and it has power to shape societies, knowledge, beliefs, values, social relations and identities. In other words, it has the exclusive power to represent things in particular ways, which are largely dependent on how language is used. Because of the media's substantively linguistic and discoursal nature, it is important to use linguistic tools in order to examine how the media chooses to represent phenomena in the world and to uncover possible prejudices in the way it does this.
In this study the representation of Muslims prior to 9/11 will be examined by analysing linguistic features and lexical patterns in the New York Times by shedding light on a small selection of linguistic features at work in discourse which might contribute to the maintenance and creation of people's worldview and stereotypes of the 'other', in this case Muslims.
The working hypothesis is that the linguistic features apparent in the analysed discourse signal a stereotyped representation of Muslims. This hypothesis is based both on the general assumptions of Critical Linguistics (CL) and Corpus Semantics (CS) and on results from earlier studies, which show that groups conceived of as the 'other', or groups that are considered not to share the ideology of the society of which the discourse is a part, are prone to be represented in stereotyped ways.
Lastly, the aim is also to test whether the approach adopted here might be used to analyse a great deal of randomly chosen data, i.e. to fi nd out whether the method can handle a quantitative analysis of the data.

1. Society
According to Fowler, Hodge, Kress and Trew (1979), the world view comes to language users from their relation to society and the institutions that it is made up of. This world view is underpinned and strengthened by a language use which is affected by the ideology of society. Consequently, the ideology in discourse is something natural to an ordinary uncritical reader who has already been socialized into his/her society's mode of thinking. There is no discourse which does not embody the world view of the society in which it is written.
The appropriateness of forms of language is established by societal factors outside the control of the language users, and the process of choosing an appropriate form of language is governed by socialisation. In other words, sociolinguistic competence is something that has been imposed on language users by society. Whenever they exercise their sociolinguistic competence, their linguistic performance is under the control of social norms. The appropriateness of a certain syntactic or lexical choice for a particular meaning is a "gift" from society and not a creation of the writer (Fowler et al 1979:194). Furthermore, this process is unconscious and although a speaker knows what is going on it might be hard to resist the pressure of social norms. In addition to these aspects, the infl uence of society seems to work in a deterministic way in that a particular social structure demands a certain linguistic variety. Social structures affect all parts of language. This means that all levels of grammar are relevant to analyse. A related aspect, which is important to recognise is that different forms of language should not be regarded as merely having a stylistic effect, but contain their own specifi c meaning. The basic assumption of CL is that one should analyse a large set of linguistic features in context and examine the aggregate ideological mediation of these features. In other words one should not look at isolated linguistic items and read off social meaning.
A fi nal caution there is no predictable one-to-one association between any one linguistic form and any specifi c social meaning. Speakers make systematic selections to construct new discourse, on the basis of systems of ideas -ideologies-and complex purposes of all kinds. To isolate specifi c forms, to focus on one structure, to select one process, in fact to lift components of a discourse out of their context and consider them in isolation would be the very antithesis of our approach. Different features and processes must be related to one another (Fowler et al 1979: 198).
In contrast to CL, although the social meaning of language is also recognised in CS, it primarily adopts quantitative methods in which the frequency of words and phrases in large corpora is thought to yield valuable information about meaning and ideological function. The method is based on corpus linguistics and the basic tool is the Concordance (or KWIC index: Key Word in Context). It searches large amounts of data for a word and displays all occurrences within a limited span of words. The evidence is then in a convenient form to be inspected for various patterns of co-occurrence (Stubbs 2001). This approach may not only be used for analysing semantic meaning of words and phrases, it can also be used to analyse constellations of repeated meanings in order to uncover ideological meanings. … [R]epresentations circulate in the social world. The world could be represented in all kinds of ways, but certain ways of talking about events and people become frequent. Ideas circulate, not by some mystical process, but by a material one. Some ideas are formulated over and over again, such that, although they are conventional, they come to seem natural. Both the representation and the circulation are profoundly affected by the mass media, which recycles the same phrases over and over again, on radio and television, and in news broadcasts, commentaries and talk shows. Corpus analysis is one way in which the propagation of phrases can be studied: both changes over time and distribution over different texts (p 149).
Stubbs criticizes CL on four main points (1995). CL is vague concerning the actual mechanism whereby language use affects habitual thought. Furthermore, CL presents no theory of how our ways of seeing the world are infl uenced cumulatively by repeated phrasing in texts. He maintains that the analyses could be strengthened by comparative and quantitative methods. Another point of criticism is "that not much is analysed" (Stubbs 1995:107) and that the selection of small samples of data is questionable as far as representativeness is concerned. Thirdly, as is expressed in the fi nal caution of Fowler et al (1978), the relation between formal features and ideology is quite unclear since on the one hand a text is perceived as a series of traces of ideological processes and on the other hand these traces are ambiguous. Moreover, this fi nal caution seems to demand analyses of whole texts and involve a great number of formal features.
These four fundamental criticisms will be addressed in this study by analysing a great deal of randomly chosen data selected from the New York Times corpus. Accordingly, the adopted view in this study is that the media has a strong infl uence on the language community it serves partly by recycling the same language several times a day, in news and current media outlets. This ubiquitous exposure to the same words and expressions "might cause us to unwittingly adopt their attitudes and opinions" (Krishnamurthy 1996:147). Below the linguistic foundation of CL and the linguistic tools identifi ed as frequent carriers of ideological meaning are presented.

Language
Having discussed the relationship between society and discourse, it is time to review how CL perceives the interaction between ideology and language and also the particular linguistic features examined in the present study.
The methodological and conceptual foundation of CL is the Systemic-Functional approach (FG) developed by M. A. K. Halliday. It rests on three basic assumptions: 1) Language serves a limited set of functions and all linguistic elements and processes express at least one of these functions. 2) The selections speakers make from the total inventory of forms and processes are systematic in nature. 3) The relationship between form and content is not arbitrary but form signifi es content (Fowler et al 1979). There are three major functions of language: the ideational function which is a means of talking about our experiences of the world, to describe processes and states and the entities involved in them, the interpersonal function which is a means of infl uencing our interlocutor's behaviour, to express our viewpoint and to elicit or change others' viewpoints and thirdly the textual function, which the other two rely on for their realization, and which is a means of forming language into coherent output. The ideational, interpersonal and textual are, consequently, functional components of the semantic system that is language (Fowler 1991b). Since form signifi es content, it is assumed in FG that a change in wording always refl ects a change in meaning. FG underpins the conceptual and methodological approach of CL to discourse. Since the common-sensical nature of discourse obfuscates ideology, one has to look at the entire meaning potential of discourse. This means that forms and processes mediate meaning in and of themselves. "The selection of one form over another points to the speaker's articulation of one kind of meaning rather than another" (Fowler et al 1979:188).
There are a number of infl uential works presenting a wide array of analytical tools that have FG as their point of departure. It is believed that one is able to read off meaning from syntax. The main linguistic features that have proven to be particularly prone to contain ideological meaning are the following; transitivity, modality, transformation, lexical classifi cation and coherence (Fowler et al 1979, Fowler 1991, Hodge and Kress 1993, Van Dijk 1998and Fairclough 2001. The following three linguistic features will be examined in the present study: lexical representation of participants, actions and syntactic structure. Lexical representation has to do with how participants, processes and objects are represented through lexical items, such as adjectives and nouns. There is great potential for representing groups in a prejudiced way by using derogatory or negatively coloured lexical labels. "The ideological character of a discourse consists in the systematic patterns and organisation of linguistic characteristics of the relevant kind, including, in particular, the systematic classifi cation of process and par-ticipants…" (Fowler et al 1978). Indeed there are different ways to conceptualise lexical representation of groups. Fairclough (1995) maintains that for instance the use of racist vocabulary can both have an ideational value by representing groups through the use of derogatory words or an interpersonal value based on the assumption that racist ideology is shared by the writer and reader. In the present study the ideational function will be focused on.
Adjectives play a fundamental role in the classifi cation of groups and the major distinction is between predicative and pronominal positions for adjectives (and other modifi ers). The fi rst is separated from the noun it qualifi es by a copular verb, whereas the latter is incorpo-rated in the noun phrase it modifi es, e.g. impeccable effects (Fowler et al 1978). Kress and Hodge (1993)  According to van Dijk (1998), one of the basic functions of group ideology is the representation of 'self' and 'others' in a polarized fashion so that positive traits are highlighted and associated with the 'in-group', whereas negative traits are associated with the 'out-group'. This general strategy of the expression of group solidarity can be conceived of as a grid in which the positive properties and actions of the 'in-group' are emphasized, the negative properties and actions of 'others' are emphasized, the negative properties and actions of the 'in-group' are mitigated and the positive properties and actions of 'others' are mitigated. The overall ideological self-interest may be expressed in the choice of lexical items that express positive or negative evaluations as well as in syntactic forms and processes (van Dijk 1998). Actions also have an important role in the representation of groups. The attribution of negative acts to the 'other' has a similar ideological function as lexical classifi cation in that the 'other' becomes closely associated with the act performed and thus creates an image of the 'other' as for instance violent. These can be strengthened by recurrently placing the 'other' in the agent role which functions to highlight responsibility. The opposite may also be achieved through de-emphasizing responsibility by placing a particular person or group as agents in passive constructions. "In this way OUR people tend to appear primarily as actors when the acts are good, and THEIR people when the acts are bad, and vice versa…"(Van Dijk 1998:33). Thus the three linguistic features analysed are closely linked to the representation of groups and people in that lexical items, such as adjectives and nouns categorise people in certain ways, negative acts can be attributed to the 'other', thus creating a representation in which certain acts become closely associated with certain groups. Lastly, the attribution of acts can be emphasized or de-emphasized through syntactic structures such as the active construction in which responsibility is highlighted and passive constructions in which responsibility is deemphasized.
In the next section we will take a closer look at how these linguistic features might be used in discourse to create a stereotyped representation of 'out groups' by reviewing works and studies that have looked at stereotypisation and discrimination in discourse.

Stereotypisation and Discrimination
Having discussed the fundamental assumptions of CL and CS regarding the relationship society, ideology and language, I will review some works that deal with how language functions to mediate discrimination and stereotypisation Discursive discrimination can be defi ned along the following points: 1) exclusion, 2) negative other-representation, 3) discriminatory objectifi cation, and 4) arguing for unfavourable treatment of groups (Boréus 2001). These points seem to mediate, in one way or another, the representation of the 'other'.
In the analysis of the European representation of the Orient in various writings, Said (1978) establishes that orientalism is a type of discourse which creates and maintains stereotypes and provides a readymade system of ideas about the Orient. This discourse is among other things made up of "certain types of statements… a manner of regularized writing, vision and study " (1978:202). Among other things, recurring collocates are an integral part of how the image of the Orient is created. Some of the words which frequently collocate are oriental sensuality and oriental despotism.
According to Krishnamurthy (1996), the wide variety of language input that we are exposed to not only helps shape our views of the language but the frequent ideas may shape our thinking. He stresses the infl uence that the media has regarding this: "Written texts have a great impact because they can be read and re-read by the consumer… the Sun has daily sales of around 3.5 million" (1996:129).
According to Said (1978) the 'other' is mainly thought of as groups and not individuals: "In newsreels or newspapers, the Arab is always represented in large numbers. No individuality, no personal characteristics or experience" (287).
A relevant aspect which is discussed in Fowler (1991) is the issue of discrimination in the media. The notion of stereotype is a central concept in understanding how discrimination works in the media. Discriminatory practices work not in terms of the individual, but in terms of some assumed group into which individuals are put, thus creating a stereotype which cultures ascribe to the group. This stereotype is applied prejudicially to individuals. The stereotype is defi ned as a set of common-sense beliefs which the culture possesses but very rarely expresses. Accordingly, groups such as young married women, teenagers, immigrants and religious groups are socially constructed concepts, "almost as fi ctitious as trolls at bridges and princesses in towers" (Fowler 1991:94). The notion of group is a tool for "handling discrimination, for sorting unequally, and it acquires much of its apparent solidity by being traded in discourse" (Fowler 1991: 94).
In van Dijk (1991), the relation between discourse and racism is discussed. Discourse plays a prominent role in the reproduction of racism and the media plays a central role in this process, since it relays both political and corporate discourse to the public, at the same time relaying its own biased perspective on ethnic affairs.
Another aspect worth studying is the lexical items used to represent minorities. Due to the fact that blatant racist accounts are very rare in western main stream media one has to analyse lexical patterns that imply a negative representation. According to van Dijk (1991) Islam and Muslims are often represented as problematic or as a threat. In other words, the lexical usage constitutes a fundamental part of this representation. Somewhere else, van Dijk (1998) maintains that: " … it is necessary to examine more systematically the semantic structure of the text for various forms of implication, indirectness… Indeed seemingly non-evaluative, non-ideological descriptions of 'facts' may imply posi-tive opinions about Us and negative opinions about Them" (Van Dijk 1998: 63).
The above claims are corroborated by Sykes (1985). In her defi nition, discrimination is not concerned with explicit hostile, stereotyped or prejudiced propositional content but rather the grammatical form in which content is mediated and the patterns of lexical choices made. She presents two small studies. In the fi rst study, she examines an article about young blacks and shows by examining transitivity, patterns of lexical choices and passivisation that other participants' responsibility in the misfortunes of young blacks are effectively obscured and mystifi ed, thus giving the impression that young blacks are primarily to blame for the social problems they are associated with in the article, e.g. "Black youths stoned the police" where black youths are placed as actors in an active construction which can be compared to a passive construction in which responsibility is de-emphasized; "The police were stoned by white youths" (Sykes 1985:86). In the second study she examines a speech from 1968 about immigrants by the British politician Enoch Powell. By looking at the lexicon, she shows that it is racially discriminatory due to the fact that it contains lexical items that would not be used by the speaker to refer to his own group. The lexical items used had strong non-human associations, e.g. "immigrant offspring" which has non-human associations. In Brown (2000), the conceptual differences and overlaps between racism and Islamophobia are examined. The similarities are that both racism and Islamophobia originate from the representation of the 'other'. Other similarities are that they are defi ned with respect to ideology and that they are constituted by prejudice. Indeed, the representation of Muslims is often made up of an amalgam of Muslims, fundamentalism, extremism and terrorism. What is meant by this is that Muslims are often represented and perceived as fundamentalists, extremists and terrorists. Accordingly, these labels seem to be an inherent part of the perception of Muslims.
Furthermore, Brown presents a list of 'closed views of Islam' from the Commission on British Muslims and Islamophobia: 1) it is seen as a single monolithic bloc, static and uninfl uenced by new realities, 2) it is seen as the 'other', not sharing any aims or values with other cultures, 3) it is seen as inferior to the West and as being barbaric, irrational, primitive, sexist, 4) it is seen as violent, aggressive, threatening, and supportive of terrorism, 5) it is seen as a political ideology, used for po-litical or military advantage. These are simply some common views of Islam and Muslims that occur in the public debate, which distinguish Islamophobia from legitimate debate, consisting of for instance disagreement with the theological bases of Islam or disagreement with what some Muslims perceive to be cultural and political consequences of Islam. Especially, three aspects associated with Islam and Muslims can be glossed from this list and these are conservatism, violence and confl ict (Brown 2000:80). In Wodak (1996) the following three thematic elements are associated with minorities: difference, deviance and threat. A more general statement made by Stubbs (2001) is that words such as extremists, fanatics, fundamentalists and militants are only used to refer to other people.

Material and Method of Analysis
The primary material used is the New York Times on CD-ROM from the years 1990, 1995 and 2000. In the 2000 version of the CD-ROM, the articles stretch from January to November. The analysed texts were fulllength articles in which the word Muslim, both as a noun and as an adjective, occurred once or more. Every tenth article containing the word Muslim was randomly selected and was saved to the Concordance. This tool isolates the word Muslim with a context of fi ve lines from the main body which was deemed suffi cient to be able to analyze the three linguistic features. This enables one to analyse a great deal of occurrences without manually having to search the whole article for occurrences of the word Muslim, which would be too laborious when dealing with so many occurrences. The word in context was subsequently transferred to Microsoft Word and analyzed manually. The number of occurrences of the word Muslim is approximately 2000 in 1990 and 1995, slightly less in 2000. The random selection generated 178, 200 and 159 occurrences of the word Muslim with a context from the years 1990, 1995 and 2000, respectively. In these were found 281 SVO-clauses, 30 passive clauses and 313 words with suffi cient context to be able to disambiguate their referents.
In the present study the following three linguistic features are examined: 1. The kind of lexical items referring to or associated with Muslims/ Muslims 1 . 2. The distribution of Muslims and other participants occurring as subjects and agents in active and passive clauses, respectively. 3. The type and distribution of actions occurring in transitive active and passive clauses with Muslims and other participants as actors.
Feature no 3 above concerns the extent to which Muslims and other participants occur in active and passive clauses as subjects or agents and the kind of actions the participants perform in transitive active and passive clauses. Whenever possible, the results were subjected to a chisquare test 2 in order to determine whether the observed distribution of analysed features are statistically signifi cant. The signifi cance level will be set at p< .05. The method of analysis is described in more detail in the next section Analysis and Results.

The level of vocabulary.
I started out by examining the lexical patterns consisting of words referring to Muslims or associated with the word Muslim. Words referring to Muslims were other nouns such as fundamentalists, rebels etc. Words associated with the word Muslim were adjectives qualifying the headword Muslim and nouns describing an event in which a Muslim participates, such as terrorist attack. The words have been divided into neutral and non-neutral ones. The classifi cation of non-neutral words is based on Brown's (2000) list of "closed views" of Islam and Muslims and van Dijk's (1991) view that Muslims are often perceived as a threat. Words that position Muslims in a context of confl ict, violence, and words that refer to Muslims in terms of groupings are defi ned as nonneutral. The latter is in accordance with Krishnamurthy (1996) andSaid (1978), i.e. the 'other' is often portrayed as a collective rather than as an individual. In short, words that contribute to the maintenance and/ or enforcement of the Muslim stereotype are considered as non-neutral. Words that have more straightforward negative denotations, such as xenophobic, violent, terrorism and militant/s lend themselves more easily to classifi cation.
The table below shows the nine most common words that all fall under the category of non-neutral words and which constitute more than one third of the total number occurrences.  Figure 1 shows the proportion of neutral and non neutral words referring to or associated with Muslims/Muslims. (1) The police gave an initial estimate of 35,000 marchers, the same number given last month for a Muslim fundamentalist demonstration (NYT 1990).
(2) Gunmen from a militant Muslim group released the wounded Prime Minister of this Caribbean island nation today, but they continue to hold other members of the Government who were taken hostage four days ago (NYT 1990).
(3) Separatists in Kashmir, the only predominantly Muslim state in largely Hindu India, want to unite the region with Pakistan (NYT 1990).

The level of grammar
Transitive clauses. The focus of the analysis is to see whether Muslims in comparison to other participants are actors in transitive active and/or passive clauses to a greater or lesser degree and which type of actions most frequently occur when Muslims compared to other participants are actors in transitive active and passive clauses. All the transitive active and passive clauses with an identifi able actor found in the data were divided as to whether the actor could be identifi ed as a Muslim or as an other participant. Accordingly, all actors that denoted non-Muslims or from the context were understood as not denoting Muslims were categorised as other participants. All in all 281 transitive active clauses were found in the data out of which 164 (58%) had a Muslim referent as actor and 117 (42%) had another participant as actor.
In table 2 we can see the distribution of Muslims and other participants occurring as subjects in transitive active clauses. It can be ideologically motivated to foreground the actor by using an active clause where the actor is in focus and acts on another entity or participant. This structure often implies responsibility, consciousness and intention. Two types of passives. There are two types of passive clauses, one in which the actor is left out completely which is often referred to as the short passive and one in which the actor comes at the end of the clause in a by-phrase referred to as the long passive (Hasselgård, Johansson and Lysvåg 1998). Both types are included and as regards the former type, the type of actor that has been left out (Muslims or other participants) has been inferred from the context. The fi rst type may serve to make causality and agency unclear and the second type may serve to take the focus away from the actor by putting the entity acted upon in the limelight. According to Fairclough: Such choices to highlight or background agency may be consistent, automatic and commonsensical, and therefore ideological; or they may be conscious hedging or deception (2001:102).
It must be stressed that within the framework of CL passive clauses with an agent are regarded as contributing to making agency and causality unclear. Sykes (1985) in particular outlines the ideological effects these might have. First it removes the agent from the prime location and thereby switches the emphasis away from the agent and onto the subject. Second it distances the agent from its action by the insertion of the particle by. Accordingly the closeness of the agent's association with its own physical act has been diminished. 30 transitive passive clauses were found with an identifi able referent of which 12 have Muslims as agents and 18 other participants as agents. Of the 30 passives 20 were short passives of which 5 had Muslims as agents and 15 other participants (see Table 4). Table 3 shows the distribution of transitive passive clauses with Muslims and other participants as agents. In passive clauses, we have other participants as agents in as many as 60% of the cases, while Muslims are agents in 40% of the cases. However, if we collapse the results in Tables 2 and 3, there is no statistically signifi cant difference at the p< 0.05 level. In other words, although Muslims to a higher extent are put as subjects in transitive clauses and to a lesser extent are put as agents in passive clauses than other participants, the differences are not great enough to rule out the possibility that the distribution is due to chance.
In Table 4 below the distribution of transitive short passive clauses with Muslims and other participants as agents is shown. However, if one only chooses to look at short passives the difference between Muslims and other participants becomes even more marked.
Other participants are put as implied agents in 75% of all short passives analysed, whereas the corresponding fi gure for Muslims is 25%. If we collapse the results in Table 4 and  (7) and other participants (8) are actors, respectively, although in the second example we have a non-fi nite construction. The last two sentences are instances of passive clauses where other participants and Muslims are agents, respectively.
(4) Last February, a Muslim group in Lebanon took responsibility for attacking a busload of Israeli tourists on the desert highway outside Cairo, killing nine of them (NYT 1990).
(5) At least 31 deaths were reported across the country today in religious riots linked to a campaign by Hindu fundamentalists to build a temple on the site of a Muslim mosque here (NYT 1990).
(6) Four people were killed and 40 wounded when shells struck Muslim districts close to the dividing line with East Beirut (NYT 1990).
(7) On Wednesday a Jewish settler was shot and seriously wounded in an attack near Ramallah in the West Bank… (NYT 1995) Type of action. As mentioned, it would be of interest to analyse the types of actions performed by Muslims and other participants in transi-tive active and passive clauses in order to determine whether these are of a violent or non-violent character.
In Table 5 are shown the number of actions performed by Muslims and other participants in transitive active clauses that are of a violent or non-violent nature. In the category violent actions, actions such as threats of violence or intention to commit violence e.g. kill, destroy and threaten are included, while the category non-violent actions includes any actions that does not imply violence. The ideological motive to highlight Muslims as actors to a higher degree than other participants was thought to be much stronger if the actions that occur in the transitive active clauses were of a violent nature and if the actions in passive clauses were of a more non-violent nature compared with other participants. The results show that Muslims carry out violent actions in 38% of the cases, whereas the corresponding fi gure for other participants is 22%. Accordingly, Muslims carry out violent actions to a higher degree when they are actors in transitive clauses than other participants. There is a statistically signifi cant difference at the p< 0. 05 level, d.f. = 1 and with a chi-square value of 8.273. Table 6 shows the distribution of violent an non-violent actions in passive sentences.  (NYT 1990) The fi rst two are examples of Muslims carrying out a violent action in a passive clause and an active clause, respectively, whereas the last two are instances of other participants carrying out a non-violent action in a passive transitive clause and a violent action in a active transitive clause, respectively.

Conclusion
The main aim of this essay was to examine the extent to which lexical categorisation and the distribution of syntactic structures contribute to a stereotyped representation of Muslims in the New York Times in three years with a fi ve-year interval.
Starting with lexical categorisation, it is apparent that the majority of the words signal a non-neutral representation of Muslims. Of course, one might claim that the choice of words agrees with the topic or event described. Many of the events involving Muslims reported in the media are violent and Muslims often act in various groupings to achieve certain goals by force or by peaceful means. So, it could be argued that the choice of words to represent Muslims, indeed, refl ects the media's goal, which is to present the truth. However, according to van Dijk (1991) topic selection is not a neutral procedure. Minorities are frequently represented in terms of very limited and stereotyped sets of topics. Muslims in particular are, in the western media, represented as problematic or as a threat to the majority culture. Indeed, in accordance with this and Browns' list of "closed views", it was found in this study that Muslims through the lexical patterns in the examined articles are associated with a limited set of contexts and circumstances, namely violence, confl icts and groups.
If we look at one of the most common words e.g. group and other words that denote a collective of people, one might say that some of the words are neutral regarding denotation and connotation. However, there seems to be a tendency to describe the "others" not as individuals who readers might identify with but as a collective (cf. Krishnamurthy 1996 andSaid 1978). Consequently, if Muslims are frequently associated with groups involved in violence and confl icts it will give rise to a stereotyped image of Muslims in which violence and confl icts are integral parts. Moreover, we can relate this to Fowler's (1991) idea that the notion of "group" in the way it is used in discourse is a tool for handling discrimination.
If we turn our attention to the syntactic choices, the results seem to indicate that Muslims are put in subject position in transitive active clauses more often than other participants and that in comparison to other participants, they occur less frequently as agents in passive clauses. Thus, this practice may contribute to a representation of Muslims as responsible and intentional actors, whereas, when it comes to other participants these aspects seem to be obscured. Together with the fact that Muslims are more often portrayed as committing violent acts, it is likely that Muslims are or become perceived as a threat. It has been a common practice in the South African media to background the responsibility of the police when acting violently upon civilians by, for instance, using the passive. According to Fairclough, this practice is an ideological as well as a linguistic process which "… assimilate[s] problematic events to preconstructed ideological frames for representing political relations in southern Africa" (1995:27).
The results also indicate that Muslims commit violent acts to a higher degree than other participants as agents in passive clauses, which thus runs contrary to my assumption that when committing violent acts references to Muslims would primarily occur as subjects in an SVOclause. However, there is no statistically signifi cant difference between Muslims and other participants, and the number of passives found and analysed are so few that it is diffi cult to generalize. This also indicates that in order to make any more conclusive claims regarding linguistic representation or ideological manifestation in discourse large sets of data that refl ect recurrent use is required, as is argued within CS.
In the light of this, the results can be interpreted as signalling a stereotyped representation of Muslims in the New York Times, or at least be interpreted as contributing to a stereotyped perception of Muslims, although, the results taken separately are not overwhelmingly conclusive. In many cases the differences are not markedly great. Nevertheless, the fact that there are differences over a relatively large set of features can be argued to be quite persuasive.
One might regard this kind of study, where random data and relatively little context have been analysed, as a test of the limits of the prevailing qualitative method adopted within the framework of CL. And it has indeed been shown that frequency and distribution of syntactic structures and lexical items have yielded quite convincing evidence that there is a pattern of stereotyped representation of Muslims.
Lastly, I leave it to future studies to further explore the limits of CL analysis as regards a great deal of randomly chosen data by including other linguistic features and by comparing Muslims with other specifi c groups, such as Christians or Jews.