Home / Articles / Volume 7 (2010) / Conversations within conversations: Intertextuality in racially antagonistic online discourse
Document Actions



Intertextuality, or the multiple ways in which texts refer to other texts, has received considerable attention in the analysis of written, especially literary, discourse (e.g., Bakhtin, 1986; Fairclough, 1992). Intertextuality is based on the view of text as a centro, a space where various discourses, motifs, and images together form a patchwork intertext that resembles a collage/montage of others’ voices (Hartman, 1992). In this perspective, the author is seen as a multidimensional space through which the utterances of others speak. Speakers, too, invoke other utterances, acts, and situations in their discourse, both in oral poetry and in everyday conversation. What renders such polyvocal discourse coherent is that intertextual relations rest on shared cultural concepts and repertoires that “serve as conventionalized orienting frameworks” (Bauman, 2004, p. 2).

Bakhtin explains the pervasiveness of intertextuality and the purpose it serves as follows:

The transmission and assessment of the speech of others, the discourse of another, is one of the most widespread and fundamental topics of human speech. In all areas of life and ideological activity, our speech is filled to overflowing with highly varied degrees of accuracy and impartiality. The more intensive, differentiated and highly developed the social life of a speaking collective, the greater is the importance of attaching, among other possible subjects of talk, to another’s word, another’s utterance, since another’s word will be the subject of passionate communication, an object of interpretation, discussion, evaluation, rebuttal, support, further development and so on. (Bakhtin, 1981, p. 337)

Intertextuality is a prominent feature of discourse on the Internet, especially in Usenet newsgroups1 devoted to controversial topics such as racism. In these groups, another’s words are frequently “the subject of passionate communication.” Evaluation and rebuttal, in particular, are common activities. Citing Wittgenstein, Zickmund (1997) characterizes “the pattern of insult and abuse” as “part of the ‘language game’ of these newsgroups,” and attributes to it “a key role in triggering a dialogue, providing a theme, or a sense of vivaciousness needed to continue a discussion” (pp. 200-201). Zickmund studied the discourse of radical hate groups on such newsgroups as alt.politics.white-power and alt.skinhead, concluding that their dialogic ‘language games’ reaffirmed their group identity based on a shared subversive ideology, at the same time as they opened up the discourse to critical participation by outsiders.

The present study looks at an ostensibly more moderate group, alt.discrimination, which was started to promote constructive discussion of racial discrimination. In fact, this newsgroup, too, is typically an arena for expression of racial supremacist ideology, but multiple ideological perspectives (e.g., those of both white supremacists and black supremacists) are represented. One of the most frequent and striking of the discourse features in this group’s exchanges is intertextuality. Participants quote one another constantly, and refer frequently to texts drawn from both within and outside the discursive universe of the newsgroup. Thus alt.discrimination constitutes a promising site in which to explore the dynamics of online intertextuality, as well as to shed further light on the broader question of the nature of racially antagonistic discourse.

This study was undertaken to address the following questions: Why and how are other texts referred to in alt.discrimination? By whom is intertextuality used, and what pragmatic purposes, if any, does it serve? How does the practice of referring to other texts work in the context of computer-mediated “hate speech?” What can the study of intertextuality in computer-mediated hate speech tell us about the attitudes of the speakers and the social groups to which they belong? The results of the research indicate that text users manipulate constructions of other’s voice in deliberate ways for specific pragmatic purposes.

I address the structure and uses of intertextual reference by analyzing one week of messages posted to alt.discrimination. Five basic types of intertextual references were coded in the data: direct references to other texts, direct quotes of other texts, inferable references to texts, references to hypothetical texts, and references to cultural texts such as common phrases or proverbs. The relation between each intertextual reference and the text producer, the text producer’s ideological position, and the text producer’s stance towards the referred to text (e.g., whether the reference is represented by the speaker as false or invalid or whether it is represented as true or valid) was measured. Quantitative analysis based on Varbrul, a software tool designed to assess the relative contributions of a number of different factors to some single, dependent variable, reveals an impressive correlation between what type of intertextual reference is used and the message poster’s stance. A participant wishing to refute another’s claim tends to quote another text directly or reference its origin explicitly in some way. When participants wish to lend authority to their own claim, however, they reference other texts without attributing them explicitly as belonging to a source, e.g., as inferable or hypothetical references. Indirect intertextual references thus function to background the source of information, thereby rendering the source less vulnerable to being discredited.

In contrast, the intertextuality strategies used to strengthen or weaken arguments have no correlation to posters’ ideologies, e.g., whether or not they are in favor of white supremacy, or whether or not they support affirmative action. This would seem to indicate that speakers’ social group or political position has no affect on the intertextual strategies they use to voice their opinions, but rather that these strategies are generally available to participants on Usenet for a variety of pragmatic purposes.

Evaluation and Hate Speech

All texts, whether spoken or written, have the meanings they do in relation to other texts within a given speech community, as well as to historically prior texts, but the relationships between texts are always constructed by text users (Thibault, 1989). It is how and why text users choose to construct this centro that is of interest in the present study. Social psychology assumes that people use language to grade some object according to an evaluative hierarchy, corresponding to some “internal” schema of opinions and attitudes, which can then be confronted with empirical evidence (Potter & Wetherell, 1987). An individual’s speech or writing indexes a considerable amount of information about the social groups to which the individual can be assigned (Thibault, 1989). Social psychology thus presumes that people use language to express their personal attitudes, opinions, and value judgments towards something “out there.”

Racially antagonistic discourse serves a self-defining social function by evaluating others—or in Zickmund’s (1997) term, the Other—in negative terms. Racially antagonistic discourse is often discussed within the broader category of hate speech. Memmi (1971) defines hate narratives as typical of colonized or oppressed people, occurring often as the step between colonized people abandoning attempts at assimilation and their actual move to revolt. This is supported by Goddall’s (1995) research into work hate narratives and Kellett’s (1995) research into rape narratives. Racially antagonistic discourse on Usenet contains colonized versus colonizer type narratives; however, the identity of the colonized group changes based on the speaker. That is, white supremacists often represent non-Whites as colonizers, and themselves as colonized, whereas non-Whites tend to adopt the opposite perspective. Racially antagonistic discourse on Usenet further differs from traditional types of oral and written hate speech in the lack of attempts in the former to use hate stratagems to expedite the audience’s coalescence into a group (cf. Whillock, 1995). The use of negative stereotypes to permit description of and action against whole groups, however, is present in both traditional and computer-mediated exchanges (cf. Burkhalter, 1999). This shared characteristic is what separates expressions of racial antagonism from the common event of “flaming” on the Internet. Flaming is typically a one-to-one activity, in which one poster targets antagonistic remarks at another poster (Kim & Raja, 1991). Racially antagonistic exchanges, in contrast, involve one poster making denigrating statements about a group of people, or about a person based on his/her perceived membership in a group, in this case a racial class or ideological agenda identified with a particular racial class.

Description of the Data

This study analyzes Usenet exchanges that took place in the newsgroup alt.discrimination. At the time this study was conducted, Usenet was the world’s largest and most widely used asychronous computer bulletin board system (Baym, 1996; Kim & Raja, 1991). Around that time, it was arranged into approximately 140 broad categories (“hierarchies”) divided and subdivided into over 14,000 individual newsgroups (Smith, 1999), of which alt.discrimination is one. A message or communication on Usenet is called a “posting” or “article” (Kim & Raja, 1991). Postings are entered chronologically, but participants may keep the same header as appeared in the earlier message to which they are referring, making it possible to keep track of the “thread” of a conversation (Collot & Bellmore, 1996). Some newsgroups are monitored by a moderator, who may restrict topics, try to keep users on topic, or restrict users from using the newsgroup (Kollock & Smith, 1996). Although the newsgroup alt.discrimination was moderated at the time of this study, the moderator seldom attempted to redirect discussions to the group’s originally stated intention (see also Lambiase, this issue).

The data corpus consists of one week of messages (N=222) posted to alt.discrimination. A posting on the Usenet bulletin board system is composed of a header, body of the message, and signature area. Most postings on Usenet are responses to previous postings, so the body typically consists of a few lines quoted from an earlier message, followed by the sender’s own comments or criticism, followed by more quoted lines and more comments, etc. Each response thus becomes a separate message, analogous to a speaking turn (Severinson Eklundh, this issue). Other people may respond to the first set of responses and others may in turn respond to their responses. Each set of responses is embedded within the previous text at whatever points the responder wishes to “speak.”2 The response sets thus mimic turn-taking within verbal conversation. There were 427 such turns in the corpus.

Each posting with embedded quotes may have any number of participants, although visual constraints on reading embedded messages limits this for practical purposes to two to three participants on average per message. The most participants involved in a set of responses or “conversation” in the present corpus was five. Each message is placed under a subtopic or thread within the overall group. There is no limit to the number of topics that may be active at any time, and participants may create a new thread in which to carry on conversations at any time. The data may be easiest to picture by equating the group or bulletin board to a file cabinet, the threads or topics to files and the postings to individual documents within the files. Each posting, like a document, is independent of all other documents but may be understood only if one reads the documents previously written or exchanged. Like documents in a file, one may also expect to find it in the file to which topic it relates or copied into multiple files if the document relates to more than one file.

One week of postings during September 1995 was downloaded from the newsgroup and saved to 36 document files, one thread per file, as identified by subject headers. The 222 postings were made up of 102 original postings and 120 responses to the original postings, where a response was defined as a message produced by means of the ‘reply’ function of the sender’s mailer system. Each response was counted separately, even if it was copied in multiple messages by subsequent speakers. The threads are identified by subject headers (all messages in a thread share a common subject line), the postings by separate items with headings consisting of the participant’s online name and the date of their posting, and the responses versus original message by system generated indentation and angle brackets preceding each line of quoted text (see Severinson-Eklundh, this issue).

A wide variety of threads were active within the one-week period on alt.discrimination, including: genetic proof of whiteness, the conviction of Abu-Jamal (an African-American man accused of killing a police officer in New York City in 1981), the morality of the death penalty, and the reunification of Germany. Sixty-five different people posted in the corpus. Very few posters identified themselves with recognizable female names, and it was not possible in most cases to determine the race of the posters. Consistent with the findings of other studies of agonistic online debate (e.g., Burkhalter, 1999; Herring, 2003), it is likely that the majority of the posters were male, but that both blacks and whites were represented.

Table 1 summarizes the number of exchanges at four levels of interaction in the data sample. Topic threads, postings, and conversational turns are defined in the preceding paragraphs. Main topics are defined in the section below on ‘issues and ideologies.’

Table 1. Summary of exchanges among 65 speakers on alt.discrimination


Although intertextuality is pervasive on the Internet, at the time this study was conducted, it had never been analyzed quantitatively.3 Thus it was necessary to devise a methodology for this purpose. I developed a coding scheme involving 11 basic types of intertextual referencing, as outlined in the following section. These are the dependent variables in the study. For independent variables, a number of different factors were hypothesized to determine choice of intertextual reference type in the preliminary stages of the research. These were subsequently reduced to the three factors that could be identified reliably enough to warrant systematic study: the identity of the individual text producer, the stance a text producer takes toward the intertextual reference being used (i.e., whether he agrees or disagrees with it), and the text producer’s ideological position with regard to race-related issues in the discussion. Accordingly, the data were coded for these three factors. Each of these is described below.

Intertextuality Types

As no established system of categorizing intertextual references was available, I catalogued all occurrences of intertextuality in the corpus using a grounded theory approach (Glaser & Strauss, 1967) and devised a classification scheme which accounted for all the variants within the data. This scheme is presented in Table 2.

Table 2. Intertextuality types

Each type may be used in conjunction with another type. Most typical is a combination of an evaluative statement (1e) and some other type. The frequencies of each intertextual type in the data sample are given in Appendix A. Type 1 references account for 51% of all intertextual references in the sample. Type 2 references make up 15%, Type 3 12%, Type 4 10%, Type 5 2%, and combined uses of two different intertextuality types make up 10% of the sample.

The types are illustrated with examples below. For the purposes of presentation, the examples have been modified from their original format in several ways. First, I inserted the name of the author of each segment for ease of identification. This information is derivable from the postings, but in the postings the actual names are not given at each segment. Second, the text has been indented to show at what level the posting was made: The original poster is indented the farthest, the second poster the next farthest, and so on. I have also inserted bolding to highlight the intertextual segment being discussed. The number(s), letter(s), and positive/negative sign which follow each example indicate the intertextuality type and stance codes that were assigned to each segment.

Type 1a: Paraphrase of texts of others

In this example, DanD summarizes poster Astro’s earlier argument, asking Astro to provide proof for his claims.

Type 1b: Reference to text of others without paraphrase of the text

In example (2), 'this' is a reference to Ben’s use of the term 'fascist', without a paraphrase of Ben’s text.

Type 1d: Reference to text generated by speaker/writer

In this example, 'what I wrote' is a reference by Pete to his previous text, although he does not summarize it.

Type 1e: Evaluative statements about texts

In this example, 'is hardly a neutral term' is an evaluation of Ben’s statement.

Type 2a: Section copied from other text

An entire paragraph was copied from Bard’s previous posting and inserted by Ronald into his posting. This coding category was only applied when the quoted segments were from third-party posters not involved in the current conversation.

Type 2b: Direct quote of word or phrase

The phrase 'slave name' is a direct quote of Ben within Theodore’s response.

Types 2c: Signature file (.sig file) quotes

The following .sig file was attached to Jeanne’s postings after her signature; it contains a quote from the late 19th-century American author Mark Twain:

Type 3: Implied reference to text

In example (8), 'Wes Cook was never a slave, nor was his father, nor even his grandfather' is implied intertextuality, as Theodore never cites a source for this information but rather asserts it as fact. Presumably, he would have obtained this information from newspapers or a genealogy of Mr. Cook’s family.

Type 1e + 4: Evaluative statement combined with a hypothetical text

In this example, the phrases 'seeks, by doing so, to express his contempt of this, our culture' and 'signifies his alienation from our civilized culture' are evaluations of a hypothetical text. Mr. Cook has, presumably, never actually said these things, but Theodore discusses them as if he had.

Type 5: Cultural texts

In the following example, James is referencing a common English-language proverb which says “if it looks like a duck (and walks/wobbles like a duck) and it quacks like a duck, it must be a duck.”

It should also be noted that because of the format of Usenet postings as described above, all the turns that are included in a current posting, but which occurred before the most recent speaker’s turns, are actually instances of intertextuality. That is, the original speaker posts a message, and each speaker in responding to the message copies the original text. The subsequent (third) poster copies the initial posting and the first responder’s comments and then inserts his own comments. This process is repeated every time a speaker responds to a posting (Severinson-Eklundh, this issue). These occurrences were not coded as intertextual references, as they are a technologically-determined feature of Usenet discourse, and thus their composition and purpose are fundamentally different from other intertextuality types.

Issues and Ideologies

Based on claims of Critical Discourse Analysts that linguistic forms of expression are manipulated to reflect the ideologies of text producers (e.g., Fairclough, 1992), it was hypothesized that choice of intertextual reference would correlate with the ideological positions of the speakers on the race-related issues discussed in the newsgroup. Twelve main issues, or supertopics (Chafe, 1994),6 were identified in the corpus. These are listed below:

1. affirmative action
2. African American behavior
3. blacks are inherently racist
4. black supremacy
5. death penalty
6. death penalty for Mumia Abu Jamal
7. Jewish behavior
8. liberals
9. Mumia was a journalist
10. white behavior
11. whites are inherently racist
12. white supremacy

Each of these 12 issues was the main topic of one or more of the 36 threads in the corpus. Because many of the same issues were addressed repeatedly, it was possible to determine how the majority of speakers felt about them. The issue concerning Jewish behavior was eliminated from further analysis because there were too few turns in which it was discussed to determine most speaker’s position on the issue. Only four of the speakers posted so infrequently that it was not possible to determine their ideological position on any of the issues. As these four speakers’ ideologies were not clear, it was not possible to correlate their ideologies and position on issues with intertextuality; accordingly, the text of these four speakers was not included in this correlation run.

Each poster was coded for their ideological position on each issue discussed. Individual speakers may be positively or negatively oriented towards an issue, or their viewpoint may be indiscernible. Accordingly, each speaker was assigned 11 issue codes, with an agree (+), disagree (-), or indeterminable (n/a) value for each. An issue may receive an indeterminable code for a given speaker if the speaker never addressed the issue or if his view point on the issue was not made clear when he did discuss it. Every intertextual reference was assigned an ideology code dependent on the position its writer takes on the 11 issues.

The 11 issues were combined into six basic ideologies depending on the grouping of speakers’ ideological positions. For example, a speaker was assigned an ideological view code “1” if he expresses an opinion in favor of affirmative action, expresses positive feelings toward what he perceives as general African American behavior, expresses no opinion on the issue of whether or not blacks are inherently racist, expresses a position of black supremacy, is against the death penalty, expresses positive feelings towards liberals and Mumia Abu Jamal’s journalism career, expresses negative feelings toward what he perceives as general white behavior, and expresses the opinion that whites are inherently racist and have no cause to believe themselves to be superior. Any other speaker expressing these same general positions receives the same code.

The six ideological groupings are shown in Table 3. Each person who falls within a particular group did not necessarily profess each of the ideologies within the group, but the ideologies he discussed agreed with the ideologies for that group. Four of the 56 speakers expressed a combination of ideologies that did not agree with the other six standard groupings or with each other; they were assigned a code indicating variation (v).

Table 3. Ideological groupings

One of these six codes was assigned to each intertextual reference, depending on the code assigned to its poster.


I further hypothesized that there would be a systematic relationship between the type of intertextuality used and the speaker’s stance toward the text to which he was referring. By stance is meant whether the speaker represents the text to which he is referring as true and valid, or as false and invalid. In order to address this hypothesis, the following scheme was developed to account for all uses of intertextuality in the data sample. This scheme and coding system identify the logically possible different stances speakers may take towards their intertextual references.

Table 4. Types of stance

Each intertextual reference was assigned a stance code depending on the stance the poster took toward the particular reference. In all, 904 tokens of intertextuality were coded by intertextuality type and stance (for exact distributions of each type, see Appendix A). In each example in this article, a stance code follows each intertextuality type code. Positive or negative stance codes can be seen after each intertextual reference in the previous examples. The following segment contains several examples of null stance codes:

Richard does not present his personal views on the veracity of Mumia’s credentials, the information or postings on the web page, or Mrs. Faulkner’s episode. He comes closest in the last case, by suggesting that it was something undesirable.

The following is an example of a positive/negative stance code:

Bro Dan describes himself as a conservative; therefore, he does not believe that “only” liberals visit prisons. The context of this communication is an argument with another conservative in which he uses the prisoners’ perspective on conservatives to support his claims, thus he does believe the resident actually feels this way.

Statistical Analysis

In addition to the codes described above, each text segment was also coded for speaker. Once the text was coded for speakers, speakers’ ideological positions, intertextual references, and speakers’ stance toward the intertextual references, a Varbrul analysis was run. This analysis was conducted to discover which, if any, factor determined what type of intertextual reference a speaker chose to use.

Varbrul is a software tool designed to assess the relative contributions of a number of different factors to some single, dependent variable (Sankoff, 1988). Varbrul weightings indicate the significant factor group calculations that result from the analysis. When a run is made, the factor that is most heavily weighted in the direction of the factor being tested is assigned the highest value, and the factor that is the least heavily weighted in the direction of the factor being tested is assigned the lowest value. 1.00 is the heaviest weight a factor can receive in favor of the test value, .50 is perfectly neutral, and .00 is the heaviest weight a factor can receive against the applied value. A negative variable run shows which factor is the most negatively affected by the dependent variable. A positive variable run shows which factor is most positively affected by the dependent variable. For example, if an intertextuality type is almost always positively used, it will score very high in a positive stance test run and very low in a negative stance test run. Values that fall between .60 and .40 are considered not strongly weighted either positively nor negatively. Two runs with Varbrul were made on the coded data, a run with negative stance as the dependent variable and a run with positive stance as the dependent variable.


Once the Varbrul analysis was completed, it was possible to determine which factors were significant and therefore which hypotheses, if any, affected intertextuality usage. Based on these calculations it emerged that although there was no significant correlation between ideology and intertextuality type, there was a significant correlation between individual speaker and stance, and an even stronger correlation between intertextuality type and stance.


None of the intertextuality types was found to correlate significantly with either the 11 issues or the six ideologies identified above. This is somewhat surprising, in that one might expect those who are arguing for a less “racist” ideology to use strategies different from those arguing in favor of “racism.” This is not to say that ideology does not affect the content of the message, but rather that ideology does not affect the strategy of intertextuality employed to express that ideology, at least in the present sample.


A number of individual speakers showed a significant correlation with certain stances (positive or negative) toward their intertextual references. However, there was no significant correlation between intertextuality type and speaker. This would seem to indicate that although speakers may prefer particular communication strategies, such as being very negative, they all have the entire range of intertextuality types at their disposal, and alternate among them.

Stance – Intertextuality Type

Varbrul revealed a significant relationship between the type of intertextual reference used and the speaker’s stance toward the intertextual reference. Certain types of intertextual references were predominantly used when the speaker represented the reference as true or valid, and other types were predominantly used when the speaker represented the reference as false or invalid. The results of this Varbrul analysis are presented in Table 5.

Table 5. Scale of significant relationship between intertextuality type and stance

The results of the Varbrul analysis thus support a significant relationship between intertextuality type and stance. The results for each intertextuality type are explained and illustrated below.

Type 3

Type 3, implied references to texts, was the most positively weighted intertextuality type in both the positive and negative run. It is the preferred intertextual reference to support the speaker’s claims. Implied or indirect textual references function to background the source of information, thereby rendering the source less vulnerable to discreditation. For example, in the following example, Richard states as “fact” that Mumia was driving a cab when the event happened but gives no source for this “fact.”

Equally revealing are cases in which direct reference in used with a positive stance. In the following example Natural uses a type 1a intertextual reference where a type 3 usage would be expected, since he is disagreeing with the source he cites.

In example (14), Natural chooses not to use a type 3 intertextual reference to support his claim, but rather cites “Rush” (Limbaugh, a conservative American radio commentator) as his source. This allows James to ignore Natural’s issues of social reform and African American behavior, and focus instead on discrediting Natural’s source. It is presumably to avoid having their sources discredited that Usenet posters use non-specific type 3 references so frequently to support their claims.

Type 1d and 1e (see examples 3 and 4)

As is to be expected, when speakers cite their own previous text, they usually represent it as valid. What is interesting is that when a reference to their previous text is combined with an evaluation of the text, the combination is used more negatively than the reference alone. Combinations of evaluation and reference to a speaker’s own text drop the positive weighting closer to, or into, the zone of neutrality. This is consistent with all other combinations of evaluations and other intertextuality types. Evaluations on alt.discrimination are used primarily to show that an argument is false or invalid.

Type 2c

Quotes in .sig files are used predominantly because the speaker feels they are valid or true. Typically they are famous quotes which support the writer’s ideology (see example 7). On occasion, however, they are famous quotes that favor their opponent’s ideology, but the writer feels that they are so blatantly false or invalid that they show their opponents’ point to be absurd. In the following example, McKinstry, who expresses white supremacist ideology, uses a quote to direct attention to the injustices perpetrated by Jews, an ethnic group that accuses others of genocide against its people.

Type 5 (see example 10)

Common phrases and proverbs are predominantly used because they have a culturally grounded meaning which goes beyond the sum of the meaning of the words (cf. Bauman, 2004), and is generally assumed by all to be true. They may, therefore, be manipulated by speakers in subtle ways to support their claims at many levels without a direct reference to this support.

Type 4

Hypothetical texts are weighted slightly to the negative but close to or in the neutral zone. These instances are frequently references to what should be said, and thus are referenced as neither positive or negative:

Type 1a and 1b (see examples 1 and 2)

References to the texts of others, frequently others involved in previous discussions, are used equally to express positive and negative stances. This causes such references to be weighted neutrally overall. When combined with an evaluative statement, however, they become strongly negative, even more negative than an evaluative statement alone. When an evaluation is combined with a direct reference (1b), it is more negative because it is both evaluating a text (as in a simple 1e) and the speaker. When a paraphrase of another’s statement (1a) is combined with an evaluative statement, it becomes the most negatively weighted type in the negative stance run, and the second most negatively weighted type in the positive run.

Type 2a

Excerpts are generally copied directly from other texts in order to pick them apart and discredit them (see example 5). The exceptions to this are when the speaker feels someone else is wrongly accusing the originator of the text and repeats the text in its entirety to prove this point.

In the above example, JR apparently believes Emily’s reference to Bard’s statement to be in error. JR inserts Bard’s earlier posting and asks Emily to explain her evaluation of it directly.

Type 2b

In the rare occurrences when a word or phrase is directly quoted within a text and is positive, it is typically when the speaker is referencing his own text. In the following example, McKinstry draws attention to his previous posting on the same issue by referencing the posting’s subject line.

More commonly, type 2b is used to discredit the arguments of others. It functions in the same way as “scare” quotes (i.e., quotation marks to signal that a word or expression should not be taken at face value), except that the words were actually used by a previous speaker. In the following example, Mark repeatedly employs this typical type 2b usage in his sarcastic evaluation of a previous poster’s text:

In short, each intertextuality type seems to serve specific pragmatic functions. The use of these intertextuality types is manipulated by speakers, according to the types’ functions, to support the speakers’ arguments and refute the arguments of others. When used effectively, intertextual references can be powerful argumentative devices. When speakers use intertextuality “incorrectly,” however, they may weaken their argument and leave themselves and their arguments vulnerable to rebuttal.

Intertextuality – Stance Alternation

A further finding is that speakers typically alternate between intertextual references they feel are valid or true and those they feel are invalid or false. When there is a sequence of intertextual references of the same stance, they are ordinarily of different types but referencing the same topic; this creates an impression of multiple references rather than one reference. The following is an example of this alternation pattern:

A linear mapping of the alternation pattern in Les and Lane’s turn exchange is as follows:

This alternating between stances invokes an atmosphere of debate, even though the participants are not actually interacting. An argument is always stronger if one is arguing against someone. Intertextual references, by their very nature, are shifts in point-of-view, invoking the presence of an interlocutor (Briggs, 1993). By shifting between points-of-view that agree with their stance and those that do not, speakers effectively create an “ideological dialectic” (Zickmund, 1997), thereby enhancing the force of their argument. This analysis is supported by the fact that posters such as Lane and Les alternate between stances rather than first stating all their positive and then all their negative stances. Indeed, there is no instance in this example of a type and stance following an identical type and stance. Speakers in a 54-turn sample alternated stances from one intertextual reference to another 51% of the time, intertextuality types 92% of the time, and either intertextuality type and/or stance 95% of the time.

The overall findings of the study are summarized in Table 6 below.

Table 6. Summary of findings

Discussion and Conclusion

That a significant relationship exists between intertextuality type and stance indicates that participants on alt.discrimination use intertextuality types strategically to accomplish particular pragmatic goals. These goals are attained by a manipulation of intertextuality types and alternation of stance. Each intertextuality type should be used in the manner that is most effective for that type, or else the speaker’s arguments may be left open for easy rebuttal. The general tendency towards alternation among intertextuality types and/or stances indicates an implicit awareness by the participants that a speaker should not be repetitive in his argumentation, lest the posting be dismissed by others as dull and redundant or, worse yet, not argumentative at all (cf. Zickmund, 1997).

These findings support the broad claim of Critical Discourse Analysts that linguistic forms of expression are manipulated to fit the social agendas of text producers (Cobb, 1994; Fairclough, 1992). Choice of intertextuality type is manipulated in order to subtly discredit another information source, or to lend it greater credibility, independent of the truthfulness of the information being referenced. The findings also reveal a debate strategy that involves heavy use of intertextual referencing to evoke “conversations within conversations.” This use appears to be a response to the fact that no interlocutor is physically present in computer-mediated discourse. By infusing their dialogues with the words of others, participants construct interlocutors with whom to agree and disagree, and thus lend greater potency to their arguments.

At the same time, the lack of significant relationship between intertextuality types and issues or ideologies shows that there is no necessary direct mapping between language use and any particular ideological position in the racially-antagonistic debates analyzed. If there are stratagems which are used by one group of participants on alt.discrimination rather than another, choice of intertextuality type is not one of them. Nor does any other rhetorical feature appear to be present in the messages of one ideological group of discussants and not in those of the other in this discourse sample, although this would need to be researched in depth before the possibility could be ruled out altogether. In the meantime, this initial research suggests that no ideological camp in online racial debates—whether the issue be black vs. white supremacy, affirmative action vs. anti-affirmative action, or racism vs. non-racism—uses linguistic strategies to build antagonism that the other side does not. This supports Zickmund’s observation that racially antagonistic discourse “obscures the distinct ideologies altogether within a single sparring language game” (p. 201). The pragmatic purpose of the speaker is what determines use of intertextual strategies rather than the speaker’s ideological position. This suggests that linguistic strategies of intertextuality are ideologically neutral, rather than markers of ideological group identity.

The findings also appear to contradict accepted perspectives in hate speech research, as found in studies such as those of Memmi (1971), Goddall (1995), and Kellett (1995). In contrast to the characterization of hate speech as practiced mainly by “colonized” groups, most of the hate speech on Usenet is posted by white males, typically considered to be the colonizers, not the colonized. How do racial hate narratives on Usenet fit into an overall understanding of the purpose of hate narratives? It is possible to imagine that these men feel colonized; this would fit into the white supremacist anti-immigration stance and anti-affirmative action position advanced by many alt.discrimination posters. They may see areas that “belong” to them, such as their privileged economic and social positions, as being invaded or already colonized by non-white others (Zickmund, 1997). The acts of violence perpetrated by skinhead groups and others would then be interpretable as a revolt against “colonizing” powers, and therefore justifiable to the perpetrators, as rape was justified in Kellett’s (1995) narratives because the rapists saw themselves as victimized by women. The suggestion that (some) white males perceive themselves as victims of society has been advanced by other researchers (e.g., Faludi, 1999), as well.

Finally, Whillock (1995) claims that people use hate speech to aid group coalescence. This claim is not supported by the present research, which found no strategies aimed at aiding group coalescence around racial hate speech on the Usenet newsgroup. Instead of trying to create a contrast between “us” and “them” as in the texts Whillock (1995) analyzed, the posters in the present study distinguished “me” from “you” and “them.” Although the “you” may be stereotyped as part of the “them” group, the posters isolate their own identity as separate from that of any other posters in the discourse. This may be because the overwhelming majority of participants in racially antagonistic computer-mediated discourse are male. Herring (1996) observed that male posters in online discussion groups are more likely than females to adopt an “opposed” stance, in which they “set themselves off as unique in the discussion, not aligned with either side but rather opposed to both” (p. 96). Alternatively, it may be a reflection of the inherently more antagonistic nature of computer-mediated communication (as suggested, e.g., by Kiesler et al., 1984). Kim and Raja (1991) and others have documented the use of “flaming,” hurling directed insults, as a frequent strategy in computer-mediated communication. Unlike flaming, however, racial hate speech on Usenet is not directed at a single person but rather at a group. Posters may direct “flames” at individuals, but expressions of racial hate are typically addressed to a group, a group characteristic, or the perception that another poster stereotypically belongs to a particular group. This practice supports Whillock’s (1995) claim that stereotyping is an integral part of hate speech. More generally, these observations underscore the benefits of studying racially antagonistic discourse in computer-mediated environments to expand understanding of hate speech.


  1. At the time this study was conducted, Usenet was the world’s largest and most widely used asychronous computer bulletin board system (Baym, 1996).

  2. The terms ‘speaker,’ ‘speaking,’ and ‘conversation’ are used in this article to refer to ‘writers,’ ‘writing,’ and ‘text,’ respectively, because these are the terms by which the posters refer to themselves and their communication.

  3. This appears to be the case still in 2010.

  4. This code did not include portions of previous messages copied into a message in the course of normal responses to those messages, as discussed at the end of the section on Methodology. Had they been included, this category would have overwhelmed the analysis, as almost all responses included copied text.

  5. A .sig file is a file created by users that automatically attaches to the end of any message they send. Typically it contains their name, occasionally their locations and personal information, and sometimes a quote from a famous (or infamous) person or written work (Kollock & Smith, 1996). It is such quotes that are coded with this notation.

  6. Chafe (1994) writes: “Supertopics achieve their coherence from the presence of some general orientation…, which extends through and supports a series of basic-level topics (our 'threads'—CHC), but exhibit no unifying scheme of their own” (p. 138).


Bakhtin, M. (1981). Discourse in the novel. In C. Emerson & M. Holquist (Eds.), The dialogic imagination: Four essays by M. M. Bakhtin. Austin: University of Texas Press.

Bakhtin, M. (1986). Speech genres and other late essays. Translated by V. McGee. Austin: University of Texas Press.

Bauman, R. (2004). A world of others’ words: Cross-cultural perspectives on intertextuality. Oxford, UK: Blackwell.

Baym, N. (1996). Agreements and disagreements in a computer-mediated discussion. Research on Language and Social Interaction, 29, 315-346.

Briggs, C. (1993). ‘I’m not just talking to the victims of oppression tonight—I’m talking to everybody.’ Rhetorical authority and narrative authenticity in an African-American poetics of political engagement. Journal of Narrative and Life History 3(1), 33-78.

Burkhalter, B. (1999). Reading race online. In M. A. Smith & P. Kollock (Eds.), Communities in cyberspace (pp. 60-75). London: Routledge.

Chafe, W. L. (1994). Discourse, consciousness, and time: The flow and displacement of conscious experience in speaking and writing. Chicago: University of Chicago Press.

Cobb, S. (1994). A critique of critical discourse analysis: Deconstructing and reconstructing the role of intention. Communication Theory, 4(2), 132-152.

Collot, M., & Belmore, N. (1996). Electronic language: A new variety of English. In S. C. Herring (Ed.), Computer-mediated communication: Linguistic, social and cross-cultural perspectives (pp. 13-28). Amsterdam/Philadelphia: John Benjamins.

Faircough, N. (1992). Discourse and social change. London: Polity Press.

Faludi, S. (1999). Stiffed: The betrayal of the American man. NY: William Morrow & Co.

Hartman, D. (1992). Intertextuality and reading: The text, the reader, the author and the context. Linguistics and Education, 4(34), 295-311.

Herring, S. C. (1996). Two variants of an electronic message schema. In S. C. Herring (Ed.), Computer-mediated communication: Linguistic, social and cross-cultural perspectives (pp. 81-106). Amsterdam/Philadelphia: John Benjamins.

Herring, S. C. (2003). Gender and power in online communication. In J. Holmes & M. Meyerhoff (Eds.), The handbook of language and gender (pp. 202-228). Oxford, UK: Blackwell Publishers.

Glaser, B., & Strauss, A. (1967). The discovery of grounded theory. Chicago: Aldine Publishing Company.

Goodall, H. L. (1995). Work hate. In R. K. Whillock & David Slayden (Eds.), Hate speech (pp. 80-121). Thousand Oaks, CA: Sage Publications.

Kellett, P. M. (1995). Acts of power, control and resistance. In R. K. Whillock & D. Slayden (Eds.), Hate speech (pp. 142-162). Thousand Oaks, CA: Sage Publications.

Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological aspects of computer-mediated communication. American Psychologist, 39, 1123-1134.

Kim, M-S., & Raja, N. S. (1991). Verbal aggression and self-disclosure on computer bulletin boards. Paper presented at the annual meeting of the International Communication Association, Chicago, IL. Washington, D.C.: ERIC Clearinghouse on Languages and Linguistics. Document no. ED334620.

Kollock, P., & Smith, M. (1996). Managing the virtual commons: Cooperation and conflict in computer communities. In S. C. Herring (Ed.), Computer-mediated communication: Linguistic, social and cross-cultural perspectives (pp. 109-128). Amsterdam/Philadelphia: John Benjamins.

Memmi, A. (1971). Dominated man. Boston: Beacon.

Potter, J., & Wetherell, M. (1987). Discourse and social psychology: Beyond attitudes and behavior. London: Sage.

Sankoff, D. (1988). Variable rules. Centre de Recherches Mathématiques. Universite de Montreal.

Smith, M. A. (1999). Invisible crowds in cyberspace: Mapping the social structure of Usenet. In M. Smith & P. Kollock (Eds.), Communities in cyberspace (pp. 195-219). London: Routledge.

Thibault, P. (1989). Semantic variation, social heteroglossia, intertextuality: Thematic and axiological meaning in spoken discourse. Critical Studies, 1(2), 181-209.

Whillock, R. K. (1995). The use of hate as a stratagem for achieving political and social goals. In R. K. Whillock & D. Slayden (Eds.), Hate speech (pp. 28-54). Thousand Oaks, CA: Sage.

Zickmund, S. (1997). Approaching the radical other: The discursive culture of cyberhate. In S. G. Jones (Ed.), Virtual culture: Identity and communication in cybersociety (pp. 185-205). Thousand Oaks, CA: Sage.

Appendix A: Frequency of Intertextuality Types

* a hyphen indicates a percentage smaller than .5

Biographical Note

Connie Hodsdon-Champeon [ cchampeon "at" biblesint "dot" org] earned her doctorate in Humanities in 1999 from the University of Texas at Arlington. She currently serves as Chief Literacy Consultant for Bibles International, working primarily in India when not in the United States.


Any party may pass on this Work by electronic means and make it available for download under the terms and conditions of the Digital Peer Publishing License. The text of the license may be accessed and retrieved at http://www.dipp.nrw.de/lizenzen/dppl/dppl/DPPL_v2_en_06-2004.html.