JCMC 4 (4) June 1999
Interactional Coherence in CMC
Program in Linguistics
University of Texas at Arlington
Table of Contents
- Interactional Coherence
- User Adaptations to the Medium
- Interactional Advantages of Text-Only CMC
- Implications for CMC System Design
- Summary and Conclusion
- About the Author
- Copyright Information
Text-only CMC has been claimed to be interactionally incoherent due to limitations imposed by messaging systems on turn-taking and reference, yet its popularity continues to grow. In an attempt to resolve this apparent paradox, this study evaluates the coherence of computer-mediated interaction by surveying research on cross-turn coherence. The results reveal a high degree of disrupted adjacency, overlapping exchanges, and topic decay. Two explanations are proposed to account for the popularity of CMC despite its relative incoherence: the ability of users to adapt to the medium, and the advantages of loosened coherence for heightened interactivity and language play.
Text-only computer-mediated communication (henceforth, CMC) has been claimed to be incoherent in various ways—fragmented, agrammatical, and interactionally disjointed—due to limitations imposed by computer messaging systems . Elsewhere I have discussed this claim as it relates to fragmentation and grammaticality in group e-mail . In this paper, I consider the claim that CMC is interactionally incoherent, and specifically, that processes of turn-taking and topic maintenance are subject to disruption and breakdown in computer-mediated environments. This claim is important in that users are more likely to engage in productive interactions (and enjoy them) using systems that facilitate, rather than hinder, interaction management. Yet because the claim appears to fly in the face of the "counter-evidence" of the enormous popularity of CMC for social interaction [4, 8], some people reject it as overly deterministic: after all, if CMC were seriously incoherent, users would not flock to the Internet so enthusiastically.
As an alternative to this view, I propose that the degree of interactional coherence characteristic of a communication medium and the popularity of that medium are essentially independent phenomena that can inform one another in complex ways. Thus in the case of turn-taking and topic development in CMC, some "incoherence" appears to be problematic for users, while other disjointed effects actually appear to attract users and enhance their enjoyment of the communication. Conversely, interactionally coherent CMC (for example, in which turn-taking is controlled) may be appreciated by users, or it may be experienced as limited and frustrating .
It is possible for CMC to be simultaneously incoherent and enjoyable because the availability of a persistent textual record of the conversation renders the interaction cognitively manageable, hence offsetting the major "negative" effect of incoherence in spoken interaction. It also opens up possibilities for other types of interaction that go beyond what is available in speech. In this as in other respects, CMC is both "less" and "more" than spoken language—or put another way, it is both dysfunctionally and advantageously incoherent.
In what follows, I first evaluate the claim that CMC is interactionally incoherent by surveying available research (including previously unpublished work) on cross-turn coherence in various text-only CMC environments, both synchronous and asynchronous. I then present evidence that users employ adaptive turn-taking strategies to overcome interactional limitations of CMC systems. The advantages of disjointed communication are then considered, including enhanced opportunities for play and intensified interactivity. I conclude by considering the implications of these observations for CMC system design, pointing out where improvements could be made while preserving interactional properties that make current systems attractive to users.
It should not be entirely surprising if computer messaging systems turn out to be as problematic as conversational environments. As many CMC historians have pointed out, computer networks were originally intended for data transmission, not as a means for social interaction [48, 50, 59; but cf. 36]. Researchers in computer-supported cooperative work have identified various limitations imposed on group interaction by the properties of the medium, ranging from high production, reception, and speaker change costs , to "chaos" resulting from the greater openness of computer-mediated systems . Two properties of the medium are often cited specifically as obstacles to interaction management:
(1)lack of simultaneous feedback, caused by reduced audio-visual cues and the fact that messages cannot overlap;
(2)disrupted turn adjacency, caused by the fact that messages are posted in the order received by the system, without regard for what they are responding to.
The first property, lack of simultaneous feedback, is a consequence of two separate features of CMC systems. First, text-only CMC is a "lean" medium [15, 16] which relies on fewer channels than face-to-face interaction for transmission of the message. Users do not see or hear their interlocutors and thus do not have access to non-verbal information about how others are responding.
Second, most multi-participant CMC systems make use of "one-way" rather than "two-way" transmission—messages are sent in their entirety when the message originator presses 'send' or 'return', rather than one keystroke at a time . One-way transmission protocols include interactive "chat" systems such as IRC (Internet Relay Chat) and MUDs (Multi-User Dimensions), along with e-mail and all forms of asynchronous computer conferencing. (Two-way protocols include split-screen "talk" systems such as UNIX "talk" and VAX "phone".) In one-way systems, it is technically impossible for the addressee to respond while the message is being written; indeed, the addressee may not even be aware she is being addressed until a complete message appears on her computer screen. Research on spoken conversational interaction shows that simultaneous feedback plays an important role in signaling listenership, timing turn-taking effectively, and maintaining continuous interaction (see McLaughlin  for a summary). Conversely, the absence of simultaneous feedback may result in discontinuity and/or overlap within turn sequences, as well as generally making it more difficult for message producers to tailor their messages to respond to recipients' interests and needs.
The second property, disrupted turn adjacency, arises from the fact that one-way CMC systems transmit messages linearly, in the order in which they are received by the system. Thus in multi-participant interactions, a message may be separated in linear order from a previous message it is responding to, if another message or messages happen to have been sent in the meantime. Delays in message transmission ("lag") as well as differential rates at which participants in different locations receive messages may even result in "speaking turns" appearing in reversed logical sequence; for example, an answer to a question may appear on the screens of some users before the question itself. The situation is different, although no less problematic, in two-way CMC systems, which isolate each participant's communication in a separate window, rather than presenting related messages as physically adjacent to one another. In contrast, spoken conversation, especially dyadic interaction, exhibits a high degree of turn adjacency; that is, relevant responses tend to occur temporally adjacent to initiating turns [51, 52]. Such "adjacency pairs" (and adjacency sequences) structure conversation and facilitate referential coherence. Conversely, when adjacency is disrupted, users may experience difficulty in tracking sequential exchanges, and interaction may become fragmented as a result.
It appears initially plausible on the basis of these observations that some degree of interactional incoherence would result—is perhaps inevitable—given the technical properties of CMC systems. Moreover, since most CMC in current use is text-only and one-way transmission, such incoherence is potentially wide-spread. However, plausibility is not the same as inevitability—to assume as much would be to fall prey to the worst abuses of "technological determinism" . Therefore we must ask: how coherent (or not) are actual computer-mediated interactions?The Evidence
One need not look far in the research literature for evidence of apparently dysfunctional interaction management in CMC. This evidence comes from descriptive studies of computer-mediated interaction making use of methods of conversation analysis (CA), as well as from research in the computer-supported cooperative work (CSCW) tradition. To the best of my knowledge, a coherent synthesis of this research has never before been attempted. In what follows, I summarize the evidence under the headings of turn-taking and sequential coherence.
Turn-taking in spoken conversation can be said to follow a normative ideal of precisely alternating turns. The word 'precisely' refers to the timing of the transition from one speaker turn to the next, which is ideally supposed to occur with no (or minimal) gap, and no overlap between speakers . 'Alternating' refers to the expectation that participants will take turns speaking in an orderly fashion; thus in dyadic exchanges, one person speaks, the other responds, the first speaks again, the second takes the following turn, and so forth. CMC, in contrast, exhibits numerous violations of both the "no gap, no overlap" principle and the principle of orderly turn alternation.
Concerning gaps, there is often a considerable time lag between when a message is sent and when it is responded to, especially in asynchronous forms of CMC such as e-mail. Days or even weeks may pass, although as some have pointed out, the gap is typically shorter than that involved in the exchange of written letters, making even asynchronous CMC seem interactive and "conversation-like" in comparison [4, 58]. Synchronous CMC involves more rapid exchanges of turns, but delays may be caused by system "lag"  and by disrupted turn adjacency, as described below.
Overlap in CMC is also problematic. On the one hand, temporal overlap in display of turns is not an option in one-way CMC, since one-way systems force messages into a strict linear order. On the other hand, overlap of exchanges is rampant in computer-mediated environments. In dyadic communication, users—unable to tell whether their interlocutor is in the process of responding or not—may become impatient and send a second message before a response to the first has been received, resulting in incomplete or interleaved exchange sequences [13, 39]. In group communication, unrelated messages from other participants often intervene between an initiating message and its response, in a likelihood proportional to the number of active participants involved in the communication [8, 37, 45].
As an illustration of the latter phenomenon, consider the following example of overlapping exchanges on a public IRC channel named #punjab (example from Paolillo ). (Line numbers were inserted by the present author for ease of reference.)
1. <ashna> hi jatt
2. *** Signoff: puja (EOF From client)
3. <Dave-G> kally i was only joking around
4. <Jatt> ashna: hello?
5. <kally> dave-g it was funny
6. <ashna> how are u jatt
7. <LUCKMAN> ssa all
8. <Dave-G> kally you da woman!
9. <Jatt> ashna: do we know eachother?. I'm ok how are you
10. *** LUCKMAN has left channel #PUNJAB
11. *** LUCKMAN has joined channel #punjab
12. <kally> dave-g good stuff:)
13. <Jatt> kally: so hows school life, life in geneal, love life, family life?
14. <ashna> jatt no we don't know each other, i fine
15. <Jatt> ashna: where r ya from?
Two extended dyadic exchanges are interleaved in this sample of chat, one between Ashna and Jatt, and the second between Dave-G and Kally. To complicate matters further, in line 13, Jatt initiates a third exchange by addressing a question to Kally. These patterns of interaction can be represented schematically as in Figure 1. (Messages generated automatically by the IRC system are excluded from this analysis.)
The perspective in Figure 1 is anaphoric—the participant lower in the diagram is considered to be responding "backwards" (or in this case, upwards) to a previous participant in each case. Dotted lines indicate interactions in which the message either initiates a new exchange with an already active participant (as in 1 and 13) or responds to a turn not included in the example (as in 3). Figure 1 shows clearly that exchanges overlap, rather than taking place in sequence—turns from one exchange regularly "interrupt" another.
Figure 1. Schematic representation of interaction in an IRC sample
Overlapping exchanges are also characteristic of asynchronous group communication, such as takes place on listserv discussion lists and Usenet newsgroups. It is not possible to reproduce a discussion list sequence of any length in this paper, since asynchronous messages are typically much longer than chat messages. However, Figure 2 presents a schematic representation of a series of consecutive messages posted as part of a discussion on a listserv discussion list named PAGLIA-L. "Exchanges" here refer to different sub-topics of discussion within the larger topic: the case of the forced early retirement of a university professor for publishing sexist views on "date rape" in his university's student newspaper. Exchange 1 concerns free speech issues versus women's rights; exchange 2 is a metadiscourse on the fact that group participants have strong opinions about the case despite having little factual information; and exchange 3 is a request for published accounts of the case and responses to that request. (For further description of the PAGLIA-L discussion, see Herring .)
The asynchronous sample also shows a high degree of overlap among exchanges: all three exchanges are "interrupted" by messages from the other two exchanges. Close parallel vertical lines in Figure 2 indicate multiple responses to a single message (e.g., message 2 received four responses in the space of this sample), and horizontal lines with multiple upward branches indicate a single message that responds to more than one previous message (e.g., message 8 responds to both messages 3 and 1). The resulting pattern of interaction is dense and complex, even for this relatively short sample of discourse.
Figure 2. Schematic representation of interaction in a discussion list sample
Nor does turn-taking in CMC adhere to the ideal that speaker turns alternate in an orderly fashion. As Figure 2 shows, there is not a one-to-one correspondence between an initiation and its response. Multiple responses are often directed at a single initiating message, and single messages may respond to more than one initiating message, especially in asynchronous CMC, where longer messages tend to contain multiple conversational moves [5, 14]. Moreover, many initiations receive no response. In a study of a social IRC channel named #yakyak, Herring and Nix  found that 18% of messages (N=226) were not responded to by other participants, counting only those messages that were clear attempts at exchange initiation. In another study of three asynchronous listserv discussions, Herring  found that more than one-third (34%) of all participants (N=117) who posted messages received no response. Message 7 in both Figures 1 and 2 are examples of unsuccessful initiations of this sort.
In the listserv study just mentioned, the vast majority (88%) of participants who were not responded to had posted only a single message, and the remaining 12% of participants had posted only two messages. In contrast, everyone who posted more than two messages was responded to. Accordingly, some participants adopt a strategy of sending out multiple messages in an attempt to attract a response. The phenomenon of "spamming"—repeatedly posting the same message—is a noxious form of this practice.
The overall frequency of multiple initiations may be seen to reflect the difficulty users have in determining others' intentions due to the paucity of feedback in text-only CMC environments. A participant cannot safely assume if he receives no response to a message that this is an intentional communication (e.g., a refusal to engage) on the part of the addressee. The addressee may not have received the message due to technical problems, may not have noticed it if she did receive it (e.g., because it was "buried" among other messages, or because she was away from her terminal when it arrived), or may have tried to return a response but failed for any number of reasons. Redundancy thus compensates for "noise" in the channel (as it does in face-to-face communication), suggesting that the computer medium is "noisy" in regard to turn-taking. Indeed, in especially active multi-participant CMC, the effect of overlap and incomplete, redundantly-initiated exchanges can be likened to a chaotic cocktail party in which every conversation is taking place, equally loudly, in the presence of every guest.
These results are found even though the data samples in the research cited above arguably exhibit a bias towards coherence. The examples represented in Figures 1 and 2 were selected to illustrate overlapping exchanges, and thus contain longer extended exchanges than might otherwise be found by taking random samples of synchronous and asynchronous CMC. The #yakyak samples were self-selected by group participants as "the best of #yakyak"; each sample is roughly "about" a single main topic. Finally, the asynchronous samples in Herring  were originally selected for analysis precisely because they appeared to "hang together" as topically-unified wholes. The fact that turn-taking is significantly disrupted even in relatively "coherent" samples such as these suggests that disruption of turn-taking is a real and persistent phenomenon in CMC.
Our discussion of turn-taking so far has focused on one-way CMC. Two-way CMC, according to McGrath [41, p. 51], "by-passes the turn taking idea" altogether, by "violating the natural communication pattern of one and only one speaker at a time". Some empirical support for this observation comes from a study of synchronous interaction using the VAX "phone" protocol conducted by Anderson, Beard and Walther . In this study, a group of three previously-unacquainted students was given a decision-making task to discuss via synchronous, two-way CMC, and they were told they would be graded on the amount of their participation. Rather than making use of their knowledge of when another participant was completing a turn so as to time turn alternations more precisely, participants tended to type all at the same time, pausing (sometimes for considerable lengths of time) as necessary to read others' contributions. As a result, the discussion exhibited a high degree of overlap (roughly 30% of all turns overlapped with another, and all three participants overlapped on more than one occasion) and a high incidence of gap, in violation of the "no gap, no overlap" ideal of spoken conversation. The researchers attribute this "dysfunctional" turn-taking to a lack of visual and auditory feedback. These findings suggest that it is not just one-way CMC, but text-only CMC in general that poses problems for the mechanics of turn-taking.
In spoken conversation, the normal expectation is that turns that "belong together"—that is, that are intended as responses or follow-ups to previous turns—will occur adjacent to one another in temporal sequence. Thus on the local level we find minimal units of conversational structure known as "adjacency pairs"  or "exchanges" , depending on whether such minimal units are thought to consist of two or three turns. On the global level, extended sequences of related turns comprise "topics" or "discourse topics" .
A corollary of this expectation is that adjacent turns will normally relate to one another, in keeping with the pragmatic principle of relevance [21, 56]. Thus an individual turn ideally relates locally to the previous turn, and globally to the discourse topic, if one has been established, except in special cases such as deliberate floutings of relevance or changes of topic. The complementary principles of adjacency and relevance I refer to here under the unified rubric sequential coherence.
Violations of sequential coherence are the rule rather than the exception in CMC. Adjacency pairs are regularly disrupted by intervening, irrelevant messages, as discussed above for one-way CMC. In example (1), Jatt's question to Ashna in message 9—"Do we know each other?"—and her answer in 14 are separated by four other messages, two of them system messages announcing another participant joining and leaving the channel, one a message between two other participants, and one from Jatt directed to Kally, none of them relevant to the exchange between Ashna and Jatt. Nor are more extreme examples hard to find—Paolillo  reports a case on the #india IRC channel of a response separated from its initiation by 50 messages! The greater the temporal and/or spatial distance between related turns, the greater the difficulty for users in identifying which previous message a current message is responding to.
In asynchronous CMC, the task of reconstructing adjacent conversational moves is additionally complicated by the fact that a single message may contain two or more moves which are physically, but not functionally, adjacent [4, 5, 14]. Thus Condon and Cech  observe that e-mail messages sent by dyads in a decision-making task sometimes contained several orientation moves, to which the other participant would respond with several suggestions, in place of the orientation-suggestion, orientation-suggestion, etc. order predicted by strict turn adjacency. This practice appears to be motivated by considerations of efficiency—by combining moves into a single turn that would normally (in conversation) be accomplished through separate turns, e-mail users reduce the number of exchanges required to complete the task and hence its "production cost" , thereby compensating for the fact that e-mail is slower than synchronous forms of communication such as "real-time" CMC and speech.
It follows from the high incidence of disrupted exchanges in CMC that turns that end up physically adjacent are often pragmatically irrelevant to one another (consider, for example, any two adjacent turns from among the first ten messages in example 1). Moreover, relevance is often weak even in sequences of turns intended to be related. In the #yakyak IRC channel, Herring  found that of all messages in three topically coherent samples, 33% violated the Gricean maxim of local relevance, and many others were only weakly relevant to the turns to which they were ostensibly responding.
An especially thought-provoking illustration of weak local relevance is the case of a robot program (or "bot") named Julia on a social MUD known as a TinyMUD, as described by Turkle . Julia was able to fool a male participant over a period of several weeks into thinking it was a human female by generating "conversational" responses when addressed, even though the computer program's responses were often of questionable relevance, as illustrated in the following exchange. (Note: the MUD system automatically encloses 'say' commands in quotation marks and prefaces them with the quotative formula '[Name] says'.)
Barry says, "Are you still holding out on me?"
Julia says, "I mean, have I convinced you I'm a computer, yet?"
Barry says, "I guess you're real...but I'd still like to get you in my DEN to run tests on you."
Julia says, "I live for your approval, Barry."
Barry says, "I'm so glad to hear that. Why don't you come over to my home."
Julia, "I want to have a child someday."
Barry says, "What a coincidence. I too wish to make a baby ...someday."
Julia says, "The date is Tue Jul 21 23:57:31, Barry."
Barry continued on with the conversation after this, apparently construing Julia's remark about having a child as flirtatious, and "her" reciting of the date and time as a joke . Although Turkle cites this exchange to illustrate a different point, the fact that exchanges of this sort pass as unexceptional is revealing for our purposes about the nature of "normal" discourse on the MUD. It suggests that users are so accustomed to the juxtaposition of irrelevant or marginally relevant turns, stemming from the indiscriminate "first-received-first-posted" feature of one-way CMC systems, that standards of local relevance in computer-mediated conversations are weakened, relative to spoken conversations in which logical adjacency is more reliably preserved.
Disruptions of extended sequences are also common. The problem of keeping track of topically-related "threads", or sequences of exchanges on a particular topic, is similar to that confronted by the user in tracking single exchanges, only more cognitively challenging. Multiple threads may become entangled, and individual threads are rarely free of disruption by irrelevant messages. In addition, keeping track of longer sequences places a greater burden on users' memories. Perhaps for these reasons, topics decay quickly in computer-mediated discussions, hastened along by off-topic digressions and tangential observations which move the discussion away from its original focus.
Empirical research on topic decay (in any communicative modality) is limited. However, some as yet unpublished research supports the perception that topics tend to decay rapidly in computer-mediated groups. In their study of a social chat channel on the Internet, Herring and Nix  found that nearly half (47%) of all turns were "off-topic" in relation to the turn to which they were responding. Moreover, the thread of discourse regularly digressed in chains of associatively-related semantic shifts, the only coherence being when the chains were loosely associated with a global topic . In the following example of a topic-shift chain, the global topic is 'jokes about blow-up dolls'. (Turns belonging to overlapping exchanges have been omitted from this example.)
14. (poosh) do those things have hair? or do you supply a wig?
15. (sigh) *snicker*
16. (blot) hair optional!
19. (poosh) blot: cool!
20. (poosh) Sinead O'Connor blow up doll!
21. (blot) lol-poosh
22. (happy1) poosh: That one would gather dust on the shelf!
23. (sigh) 9 ball, side pocket!
27. * blot is behind 8 ball!
28. * poosh is under the table
In this sequence, poosh shifts from the idea of blow-up dolls with no hair (lines 14-19) to the idea of dolls modeled after bald people, in this case, Sinead O'Connor, a popular singer who used to shave her head (line 20). From there, sigh jumps in line 23 to the idea of pool balls, blot in line 27 shifts from the idea of pool balls as bald to a specific pool ball and his (metaphorical) position in relation to it, and poosh shifts from the idea of blot's position to her own position in relation to the pool table. Through these tenuously-associated parallel shifts (cf. Hobbs ), the topic digresses rapidly within the space of a few turns.
This digression can be represented graphically as in Figure 3. In order to produce this diagram, each turn was numbered vertically, and distance from the previous functionally-related turn was plotted horizontally for each turn on a scale of zero to four. The values of the scale were defined as follows: 0=same topic, 1=new association within locally established script, 2=new association within globally established script, 3=related but introduces a new script, 4=not clearly related, introduces a new script. Thus the topical distance is 1 between turns 14 and 16 (the new idea of 'optionality' is introduced into the local script about hair on blow-up dolls), is 3 between turns 16 and 20 (the new but related script 'blow-up dolls modeled after famous women' is introduced), is 4 between turns 20 and 23 (the 'pool game' script is not clearly related to 'blow-up dolls'), and so forth.
Figure 3. Schematic representation of topic drift in an IRC sample
(T='on topic'; P='parallel shift')
In an ideal topically-coherent discussion, the majority of moves would be on-topic or single-step parallel shifts with frequent returns to the global topic . Such a discussion would have a predominantly vertical orientation when diagrammed, in contrast to the horizontal orientation displayed in Figure 3.
Asynchronous discussions also tend towards topic decay. In a study of an Internet discussion list devoted to the 1995 Oklahoma City bombing, Lambiase  found that during the first nine days of the list's existence, the percentage of messages on the group's global topic decreased steadily from 65% to 33%. The list met an untimely demise less than a month after it started, Lambiase maintains, largely as a result of disputes over topic management. A tendency for global topics to fragment into competing subtopics was also found for three asynchronous listserv discussion groups studied by Herring .
To be sure, topical fragmentation over time is also characteristic of spoken multi-party conversation. However, the tendency is exacerbated in one-way group CMC due to a lack of feedback. Participants compose responses to a topic simultaneously without knowing what (or even that) others are writing; as a result, multiple competing new directions for discussion are introduced. As participants respond to others' responses, the chances that they will move further away from the original topic—and that new topics of discussion will arise—increase exponentially. As Lambiase's study shows, even when computer-mediated groups consciously place a high value on keeping on-topic, "on-topicness" is difficult to enforce.
From the above litany of difficulties concerning interaction management in CMC, one might wonder that users succeed in—and find any enjoyment in—interacting via computers. Yet both synchronous and asynchronous CMC are enormously popular, and users regularly ascribe to both a conversational, interactive quality. There are two possible reasons for this popularity. The first is that users successfully accommodate so as to be able to manage interaction effectively, despite the limitations of CMC systems. The second is that the "limitations" are in some respects advantageous: the loose, fragmented nature of computer-mediated interaction constitutes part of its appeal, and may even lead some users to prefer it to spoken interaction. In what follows, I present evidence to support both of these views.
User Adaptations to the Medium
Certain interactional features of CMC are best understood as adaptive to the medium. Users innovate alternative methods of signaling listenership and negotiating turn alternation, despite the absence of traditional forms of feedback. They also employ alternative means of signaling connection across turns, in an attempt to offset the tracking problems caused by disrupted adjacency. Some of these adaptations are discussed below.
Contrary to the claim of some CMC researchers [5, 20, 58] that minimal responses are rare in CMC, post-turn minimal responses or 'backchannels'  proliferate in some synchronous CMC environments. In a social MUD studied by Cherny , back-channels are so frequent that the MUD participants have devised a set of programming "shortcuts" to type the most common of them, which include 'nods', 'giggles', 'hehs', 'grins', 'smiles' and '?' (questioning look) (p. 186). Even in asynchronous groups, where it is widely considered a breach of 'netiquette' to send messages that contain no significant original content [24, 44], the quoting strategy described below enables respondents to insert short responses to individual points from previous messages. Jacobs-Huey  cites an example from an African-American listserv in which a woman quotes an extended exchange between two other women for the sole purpose of inserting supportive backchannels ("Yes, break it down, Sister"; "Amen!"; "Well said") after the comments of one of the women. Practices such as these heighten the interactivity of CMC by signaling listenership and encouraging others to continue.
Other adaptive practices facilitate the timing of turn alternation, or in the terminology of Clark and Brennan , reduce speaker change cost. One IRC user reported to the author that he and a regular IRC conversational partner had devised the practice of typing a "%" symbol at the end of a message to indicate that they were not yet ready to give up the floor, and thus that the other should wait before taking the next turn. In one-way CMC where the norm is one message=one turn, this practice reduces the incidence of overlapping exchanges. Similarly, Cherny [8, p. 181] describes a MUD meeting in which participants "raised their hands" (i.e., by using a command to describe their character as raising its hand) and were called on by a moderator, as a means to facilitate focused discussion.
In contrast, in the two-way VAX 'phone' study described above , users adopted a strategy which increased the amount of overlap. While initially appearing dysfunctional, the overlap-pause strategy observed by Anderson et al. can be viewed as a more efficient form of turn-taking than the spoken language model of precisely alternating turns. This is due to the fact that typing is slower than reading: multiple participants typing at the same time, then reading each other's messages, takes less time (and keeps all participants more consistently engaged) than one person at a time typing while the others wait. CMC makes possible this more efficient form of turn-taking because of the availability of a persistent textual record of the preceding interaction. In speech, it is not feasible for everyone to talk at the same time because participants cannot simultaneously produce utterances and process the utterances of others. In CMC, however, everyone's contributions are recorded as text on the screen, available to be read and reread as necessary until they have been fully processed.
Specific devices have also arisen in response to the need to reference the previous message to which one is responding under conditions of disturbed adjacency. In multi-party synchronous modes, it is conventional to preface a turn with the user name (or nickname) of the intended addressee; this practice is termed addressivity by Werry . Its usefulness as a turn-tracking device is evident in example (1) from the IRC channel #punjab. Without the practice of addressivity, it would be difficult to determine who is responding to whom in that sample.
In asynchronous group discourse, different strategies have similar tracking functions. Linking is the practice of referring explicitly to the content of a previous message in one's response, as for example when a message begins, "I would like to respond to Diana's comment about land mines". A high percentage of Usenet and listserv messages include a linking move of this sort [4, 25]. Quoting, or copying portions of a previous message in one's response [53, 54], often functions as a subtype of linking, as in the following example from a soap opera fan newsgroup (example from Baym ). The name and e-mail address of the person quoted is given in a system-generated 'pointer' line that precedes the quote; the system also prefaces each line of quoted material with a special symbol (in this case, an angle bracket):
janed@ABC.bigtel.com (Jane Doe) writes:
>I can't believe how horrible Natalie looks. Has she put on a lot of weight?
I agree, but she has always had a somewhat round face, so if she did put on weight,
I think that would be accentuated.
Quoting creates the illusion of adjacency in that it incorporates and juxtaposes (portions of) two turns—an initiation and a response—within a single message. When portions of previous text are repeatedly quoted and responded to, the resulting message can have the appearance of an extended conversational exchange [32, 53]. These practices are adaptive in that they allow interactants to maintain and track patterns of turn-taking despite overlapping exchanges and delayed responses.
Last but not least, a concern for topical coherence is reflected in the organization of asynchronous discussion groups on the Internet. Listserv lists and Usenet newsgroups are defined around a global topic—as reflected, for example, in group names such as the Linguist List and soc.culture.india—that indicates in a general way what kinds of discussion topics can appropriately be raised in the group. The practice of 'threading' on Usenet, or sorting messages into a 'thread' based on the fact that they share a common subject line [4, 34], defines discussion topics, albeit in a mechanistic way. In addition, discussion groups often have 'moderators' or 'listowners' whose functions may include seeing to it that discussion remains "on topic" . These structural mechanisms mitigate the tendency towards topical fragmentation in extended exchanges.
Interactional Advantages of Text-Only CMC
Despite user adaptations, however, text-only CMC remains loosely coherent in comparison with the interactional norms for face-to-face conversation. This is especially apparent in CMC used for recreational purposes, for example in social MUDs and IRC channels, as well as many Usenet newsgroups. Indeed, in some groups, incoherence may be the norm rather than the exception (see, e.g., example 2). Yet clearly, given the popularity of Usenet, IRC, and MUDs [4, 48, 49], reduced interactional coherence is not a serious impediment to users' enjoyment of recreational CMC. Not only that, it might be an advantage.
Relaxed norms of coherence can be liberating, giving rise to increased opportunities for language play. Danet, Ruedenberg-Wright & Rosenbaum-Tamari  claim that CMC is "an inherently playful medium", citing as an example an IRC "party" in which participants textually enacted the activity of "smoking dope". The associative topic shifts in IRC example (3) above are clearly intended to be humorous; collaborative joking sequences of this sort (what Chiaro  calls "joke chains" in spoken conversation) are well attested in both synchronous and asynchronous CMC.
Consider further the following asynchronous example cross-posted to several Usenet newsgroups, which makes use of three layers of quoting (indicated by the number of angle brackets preceding each quoted line) to achieve a "chained" effect:
On 30 Jan 1998 00:42:15 GMT, email@example.com (Tlerll) wrote:
>In article <firstname.lastname@example.org>,
>Frank Rizzo <email@example.com> wrote:
>>> This means ABSOLUTELY NOTHING TO ANYONE EXCEPT ME.
>>then why the need to post
>and why the need to complete sentenc
so why the need to verbs.
This sequence, like the IRC sequence in example (3), exploits the tendency in CMC towards loosened local relevance, in that the last two "turns" are not strictly relevant to what comes before. It is not surprising that a weakening of relevance norms would invite humorous play. Jokes violate Gricean conversational maxims by definition [2, 47], and relevance is the most basic of Grice's four maxims . Example (5) also illustrates another property of humor in CMC, which is that it is often metalinguistic in nature, taking as its object CMC itself. Example (5) is as much a joke about endlessly embedded quotes in e-mail messages, as it is about the original comment which triggered the sequence. Similarly, Danet, Wachenhauser, Bechar-Israeli, Cividalli, and Rosenbaum-Tamari  describe a comic IRC theatrical production, "Hamnet", which plays off of the communicative conventions of IRC, and Paolillo  has analyzed a Usenet newsgroup, alt.religion.kibology, whose subject matter consists entirely of satire of discourse taking place elsewhere on Usenet.
On-line surveys reveal that humorous messages are the most highly appreciated types of messages in computer-mediated environments, even in those one would not necessarily characterize as "recreational", for example, academic discussion groups [3, 24]. The potential for humorous play inherent in the (interactionally less coherent) computer-mediated environment constitutes one of the biggest attractions of CMC for many users.
Another notable attraction is the ease with which group CMC involves users in multiple, simultaneous exchanges. Jatt in example (1) and "B" in example (2) are participants who engage in parallel interactions within a single discussion. Turkle  sees this as a distinct advantage of the medium. She interviewed young adults who report participating in four different MUDs simultaneously, each in a different window on their computer screen. In each MUD, they "play" a different character (with different personalities, genders, and sometimes even different species), all of whom may be actively engaged in interaction at any given moment. Like the phenomenon of 'multi-tasking' (working on different projects simultaneously in different windows on one's computer screen), participation in multiple exchanges maximizes stimulation and minimizes opportunities for boredom; that is, it reduces receiver cost . CMC thus enables a greater intensity of interaction—that is, more interactions concentrated within a single temporal frame—than is possible face-to-face. To appreciate this, one has only to compare a typical IRC session with a face-to-face social event, such as a party, to realize the limits on multiple simultaneous interaction in the latter.
Viewing CMC as more socially desirable than analogous face-to-face interaction—in Walther's  term, as 'hyperpersonal' rather than as 'interpersonal' interaction—helps to explain the popularity and the conversational "feel" of CMC despite the numerous apparent shortcomings of CMC systems as conversational environments. Reduced feedback and loosened adjacency enable a qualitatively different kind of interaction from that possible in spoken conversation, and this contributes to CMC's 'hyperpersonal' appeal. Thus the "attractions" of CMC described in the preceding paragraphs can be seen as the flip side of the "incoherence" coin—loose inter-turn connectedness and overlapping exchanges have both advantages and disadvantages, depending on the purposes for which users engage in computer-mediated interaction.
It would be difficult to realize the playful and multi-interactional possibilities of CMC systems were it not for the availability of a persistent textual record of the interaction in most forms of text-only CMC. E-mail leaves a record which must be actively deleted by the user in order to dispose of it, and many listservs archive list discussions. Even in synchronous CMC such as IRC and MUDs, where logs are not automatically kept by the system, users still have the benefit of seeing the immediately preceding interaction preserved on their computer screens for as long as it takes for new messages to cause the old ones to scroll up and off the screen. In this respect, even the least persistent synchronous interface is more persistent than spoken language, which disappears immediately once it is uttered.
Persistent conversation aids the user's cognitive processing. The predilection towards meta-humor and meta-play in CMC can be attributed in part to the fact that CMC persists as text on a screen and is subject to conscious reflection in ways that spoken language is not, thereby facilitating a heightened meta-linguistic awareness . Users are able to participate in simultaneous multiple interactions without getting hopelessly lost or confused, because there is a typed record to which they can refer to keep track of what is going on. Without textual persistence—if, for example, messages disappeared from the screen immediately after they were read—CMC would no doubt be more interactionally incoherent, and a great deal more limited in its uses.
Implications for CMC System Design
The above observations suggest a number of ways in which CMC systems could be designed to improve interactional coherence, as well as pointing to features of "incoherence" it would be desirable to maintain. In this section, I identify three general desiderata for system design. First, logging/archiving capabilities should be enhanced, in light of the crucial role played by persistent text in facilitating cognitive processing. This is especially needed for bulletin board systems (such as Usenet) and synchronous forms of CMC which currently preserve textual records for limited periods of time and (in the case of Usenet) in a format which makes it difficult to reconstruct the actual sequence in which messages were received. The focus in such enhancement should be on making logs available to the users in a presentational format that is easily accessible during participation in computer-mediated exchanges. (Some more specific suggestions for how this might be accomplished are advanced below.)
Second, enhancements should be made to reduce incoherence (disruptive overlaps, uncertainly as to interlocutor's intent, etc.) caused by insufficient feedback. Message transmission should ideally be two-way, so as to enable simultaneous feedback during message production. This would entail overcoming two additional problems. Two-way systems (such as UNIX 'talk') are currently difficult to log, in that each participant's words scroll independently in a separate "window" or section of the screen; without access to temporal information, it is difficult to reconstruct after-the-fact the relationship between the utterances in the separate windows. One solution to the logging problem would be to log the interaction as a QuickTime movie, and provide an interface with a scroll bar that would allow users to play back the movie, pausing it where they wished in order to refresh their memories on what was said previously.
The second problem concerns the feasibility of adapting two-way transmission for multi-participant interaction on a scale similar to that currently enjoyed by one-way modes such as IRC and MUDs. Each participant in the 'talk' and 'phone' protocols is reserved a separate, and permanent (for the duration of their connection) space on the screen, with screen size effectively limiting the number of individuals who can participate at one time. A possible approach to this limitation is to reduce the size of the space allotted each participant (allotting to each, say, a limit of two lines), in conjunction with logging the interaction as a QuickTime movie as described above. However, this would still limit the number of participants who could be accommodated on a single screen, as well as having the undesirable effect of limiting users' access to the text of earlier interaction. An alternative to assigning connected participants a "permanent" space would be to display them only when they start to type, as well as for some interval of time after they have finished typing, in order to allow others to process their message. Such an interface would have a potentially limitless number of participants fading on and off the screen, sometimes overlapping (with messages in different stages of completion), sometimes in sequence, depending on the patterns of interaction of the group. A separate window or sidebar display could be provided containing a list of all individuals currently connected to the group, so that users would know who was available to be addressed.
The third and final desideratum is for designs which reduce the number of incoherent sequences by facilitating tracking and linking of logically connected turns. Here there is a need to make available to the user different views of the interaction, e.g., one that is linear (as in current one-way modes) so that the actual sequence of events can be reconstructed, and another that physically groups related messages adjacent to one another, so that distraction from irrelevant intervening messages is reduced. The linear, sequential presentation is easy to produce, since it requires no knowledge by the computer other than the order in which messages were received by the system. Linking features could be incorporated into a linear presentation that allows users to designate which message(s) their message is a response to; the two (or more) messages would then appear on the screen highlighted in the same color (or connected by a line) for a period of time during which they would be "active", with the link then fading so as not to compete visually with more current active links.
The non-linear presentation, which should be accessible as a separate screen, is more problematic. I visualize it potentially taking the form of topic chains or trees, the advantage of the latter being that it can represent hierarchical relations among concepts (via higher and lower "nodes") and indicate subtopics as branches of the tree. However, such a design would require either a system with sophisticated analytic capacities—since it would have to determine hierarchical relations among turns based on their content—or require users themselves to designate their contributions as relevant to one or the other (or perhaps as initiating a new) topic tree. Both of these solutions are problematic, it seems to me, if they must be performed in real time (i.e., as each individual message is posted), rather than at the completion of a sequence, from which perspective a hierarchical analysis is easier to perform. I leave this as a problem for future research.
Summary and Conclusion
In this paper, I have presented evidence that interactive exchanges in a variety of CMC modes tend to be less tightly stitched together than in face-to-face conversation: responses are often separated from the turns they are responding to, topics tend to decay quickly, and multiple, overlapping exchanges often share the same channel. On the one hand, this creates potential confusion that users seek to minimize by adopting compensatory strategies. On the other hand, some users exploit the potential of loosened coherence for the purposes of play and to enjoy intensified interactivity, especially in synchronous modes. These uses, which in some respects extend the limits of what is possible in spoken conversation, are facilitated by the availability of a persistent textual record of computer-mediated interaction.
These findings have implications for system design, as well as for the broader question of the extent to which the physical properties of CMC technologies determine how human beings interact when using them. Technological determinants can not be dismissed as irrelevant in the case of interactional coherence—they have clear effects. What will perhaps come as a surprise to some technological determinists, however, is that these effects are not devastating; indeed, they comport some unexpected advantages. Interactional coherence in CMC thus emerges as a complex phenomenon which cannot be equated in any simple way with the popularity (or users' enjoyment) of CMC systems.
 J. F. Anderson, F.K. Beard and J.B. Walther, "The Local Management of Computer-Mediated Conversation", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 S. Attardo, Linguistic Theories of Humor, Mouton de Gruyter, Berlin, 1994.
 N. Baym, "The Performance of Humor in Computer-Mediated Communication", Journal of Computer-Mediated Communication 1(2), 1995.
 N. Baym, "Agreements and Disagreements in a Computer-Mediated Discussion", Research on Language and Social Interaction 29(4), 1996, pp. 315-345.
 S. Black, J. Levin, H. Mehan, and C. Quinn, "Real and Non-real Time Interaction: Unraveling Multiple Threads of Discourse", Discourse Processes 6, 1983, pp. 59-75.
 C. B. Cazden, "Play with Language and Meta-linguistic Awareness: One Dimension of Language Experience", Play—Its Role in Development and Evolution, J.S. Bruner (Ed.), Penguin, New York, pp. 603-608.
 W. L. Chafe, Discourse, Consciousness and Time. Chicago: University of Chicago Press, 1994.
 L. Cherny, Conversation and Community: Chat in a Virtual World, CSLI Publications, Stanford, 1999.
 D. Chiaro, The Language of Jokes: Analysing Verbal Play, Routledge, London, 1992.
 N. Cho, "Linguistic Features of Electronic Mail", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 H. H. Clark and S.E. Brennan, "Grounding in Communication", Perspectives on Socially Shared Cognition, L.B. Resnick, J.H. Levine, and S.D. Teasley (Eds.), American Psychological Association, Washington, pp. 127-149.
 S. L. Condon and C.G. Cech, "Functional Comparisons of Face-to-Face and Computer-Mediated Decision Making Interactions", Computer-Mediated Communication: Linguistic, Social and Cross-Cultural Perspectives, S. Herring (Ed.), John Benjamins, Amsterdam, 1996, pp. 65-80.
 S. L. Condon and C.G. Cech, "Discourse Management Strategies in Face-To-Face and Computer-Mediated Decision Making Interactions", Electronic Journal of Communication/La revue électronique de communication 6(3), 1996.
 S. L. Condon and C.G. Cech, "Discourse Management in Three Modalities", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 R. L. Daft and R. H. Lengel, "Information Richness: A New Approach to Managerial Behavior and Organization Design", Research in Organizational Behavior, volume 6, B.M. Staw and L.L. Cummings (Eds.), JAI Press, Greenwich, CT, 1984, pp. 191-233.
 R. L. Daft and R. H. Lengel, "Organizational Information Requirements, Media Richness, and Structural Determinants", Management Science 32, 1986, pp. 554-571.
 B. Danet, T. Wachenhauser, H. Bechar-Israeli, A. Cividalli, and Y. Rosenbaum-Tamari, "Curtain Time 20:00 GMT: Experiments With Virtual Theater on Internet Relay Chat", Journal of Computer-Mediated Communication 1(2), 1995.
 B. Danet, L. Ruedenberg-Wright and Y. Rosenbaum-Tamari, "Smoking Dope at a Virtual Party: Writing, Play and Performance on Internet Relay Chat", Network and Netplay: Virtual Groups on the Internet, S. Rafaeli, F. Sudweeks, and M. McLaughlin (Eds.), AAAI/MIT Press, Cambridge, MA, 1997.
 T. L. Fanderclai, "Like Magic, Only Real", Wired_Women: Gender and New Realities in Cyberspace, L. Cherny and E.R. Weise (Eds.), Seal Press, Seattle, 1996, pp. 224-241.
 A. Georgakopoulou, "On for Drinkies? E-Mail Cues of Participant Alignments", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 H. P. Grice, "Logic and Conversation", Syntax and Semantics vol. 3: Speech Acts, P. Cole and J.L. Morgan (Eds.), Academic Press, New York, 1975, pp. 183-198.
 K. Hafner and M. Lyon,Where Wizards Stay Up Late: The Origins of the Internet, Simon & Schuster, New York, 1996.
 C. Hale, Wired Style: Principles of English Usage in the Digital Age, HardWired, San Francisco, 1996.
 S. C. Herring, "Politeness in Computer Culture: Why Women Thank and Men Flame", Cultural Performances: Proceedings of the Third Berkeley Women and Language Conference, M. Bucholtz, A. Liang, L. Sutton, & C. Hines (Eds.), Berkeley Women and Language Group, 1994, pp. 278-294.
 S. C. Herring, "Two Variants of an Electronic Message Schema", Computer-Mediated Communication: Linguistic, Social and Cross-Cultural Perspectives, S. Herring (Ed.), John Benjamins, Amsterdam, 1996, pp. 81-106.
 S. C. Herring, "Humor that Binds: Joking and Coherence in Internet Relay Chat", Paper presented at the International Society for Humor Studies conference, University of Oklahoma, July 13, 1997.
 S. C. Herring, "Le Style du Courrier Électronique: Variabilité et Changement", Terminogramme 84-85, 1998, pp. 6-14.
 S. C. Herring, "The Rhetorical Dynamics of Gender Harassment On-Line", The Information Society, special issue on The Rhetoric of Gender in Computer-Mediated Communication, Laura J. Gurak (Ed.), In press.
 S. C. Herring, "Who's Got the Floor in Computer-Mediated Communication? Edelsky's Gender Patterns Revisited", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 S. Herring and C. Nix, "Is "Serious Chat" an Oxymoron? Academic vs. Social Uses of Internet Relay Chat", Paper presented at the American Association of Applied Linguistics, Orlando, FL, March 11, 1997.
 J. R. Hobbs, "Topic Drift", Conversational Organization and Its Development, B. Dorval (Ed.), Ablex Publishing, Norwood, NJ, 1990, pp. 3-22.
 C. B. Hodsdon, "Conversations within Conversations: Intertextuality in Racially Antagonistic Dialogue on Usenet", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 L. Jacobs-Huey, "...BTW, how do YOU wear your hair? Identity, Knowledge and Authority in an Electronic Speech Community", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 P. Kollock and M. Smith, "Managing the Virtual Commons: Cooperation and Conflict in Computer Communities", Computer-Mediated Communication: Linguistic, Social and Cross-Cultural Perspectives, S. Herring (Ed.), John Benjamins, Amsterdam, 1996, pp. 109-128.
 J. J. Lambiase, "Hanging by a Thread: Topic Development and Death in an Electronic Discussion of the Oklahoma City Bombing", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 J. C. R. Licklider, R.W. Taylor, and E. Herbert, "The Computer as a Communication Device", Science and Technology: For the Technical Men in Management, April, 1968, pp. 21-31.
 W. Lunsford, "Turn-Taking Organization in Internet Relay Chat", Unpublished ms, University of Texas at Arlington, 1996.
 M. L. Markus, "Finding a Happy Medium: Explaining the Negative Effects of Electronic Communication on Social Life at Work", ACM Transactions on Information Systems 12(2), 1994, 119-149.
 L. E. Marvin, "Spoof, Spam, Lurk and Lag: The Aesthetics of Text-Based Virtual Realities", Journal of Computer Mediated Communication 1(2), 1995.
 J. C. McCarthy, V.C. Miles, A.F. Monk, M.D. Harrison, A.J. Dix, and P.C. Wright, "Text-based On-line Conferencing: A Conceptual and Empirical Analysis Using a Minimal Prototype", Human-Computer Interaction 8, 1993, pp. 147-183.
 J. E. McGrath, "Time Matters in Groups", Intellectual Teamwork: Social and Technical Foundations of Cooperative Work, J. Galegher, R.E. Kraut, and C. Egido (Eds.), Lawrence Erlbaum, Hillsdale, NJ, 1990, pp. 23-61.
 A. McKinlay, R. Procter, O. Masting, R. Woodburn, and J. Arnott, "Studies of Turn-taking in Computer-Mediated Communications", Interacting With Computers 6(2), 1994, pp. 151-171.
 M. L. McLaughlin, Conversation: How Talk Is Organized, Sage Publications, Beverly Hills, 1984.
 M. L. McLaughlin, K.K. Osborne and C.B. Smith, "Standards of Conduct on Usenet", Cybersociety: Computer-mediated Communication and Community, S. Jones (Ed.), Sage, Thousand Oaks, CA, 1995, pp. 90-111.
 J. C. Paolillo, "Joke Newsgroups on Usenet", Paper presented at the International Society for Humor Studies conference, University of Oklahoma, July 13, 1997.
 J. C. Paolillo, "Conversational Codeswitching on Usenet and Internet Relay Chat", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 V. Raskin, Semantic Mechanisms of Humor. Dordrecht: D. Reidel Publishing Co., 1985.
 E. M. Reid, Electropolis: Communication and Community on Internet Relay Chat. Senior Honours thesis, University of Melbourne, Australia, 1991.
 E. M. Reid, Cultural Formations in Text-Based Virtual Realities, master's thesis, University of Melbourne, Australia, 1994.
 H. Rheingold, The Virtual Community: Homesteading on the Electronic Frontier, Addison-Wesley, Reading, MA, 1993.
 H. E. Sacks, E. Schegloff, and G. Jefferson, "A Simplest Systematics for the Organization of Turn–Taking for Conversation", Language 50, 1974, 696–735.
 E. A. Schegloff, "Sequencing in Conversational Openings", American Anthropologist 70, 1968, pp. 1075-95.
 K. Severinson Eklundh, "To Quote Or Not To Quote: Setting the Context for Computer-Mediated Dialogues", Computer-Mediated Conversation, S. Herring (Ed.), Forthcoming.
 K. Severinson Eklundh and C. Macdonald, "The Use of Quoting To Preserve Context in Electronic Mail Dialogues", IEEE Transactions on Professional Communication 37 (4), 1994, pp. 197-202.
 J. Sinclair and M. Coulthard, Towards an Analysis of Discourse, Oxford University Press, 1975.
 D. Sperber and D. Wilson, Relevance: Communication and Cognition, Harvard University Press, Cambridge, MA, 1986.
 S. Turkle, Life on the Screen: Identity in the Age of the Internet, Simon & Schuster, New York, 1995.
 L. Uhlírová, "E-mail as a New Subvariety of Medium and its Effects upon the Message", The Syntax of Sentence and Text: A Festschrift for Frantisek Danes, S. Cmejrková and F. Stícha (Eds.), John Benjamins, Amsterdam, 1994, pp. 273-282.
 J. B. Walther, "Computer-Mediated Communication: Impersonal, Interpersonal and Hyperpersonal Interaction", Communication Research 23(1), 1996, pp. 3-43.
 C. C. Werry, , "Linguistic and Interactional Features of Internet Relay Chat", Computer-Mediated Communication: Linguistic, Social and Cross-Cultural Perspectives, S. Herring (Ed.), John Benjamins, Amsterdam, 1996, pp. 47-63.
 S. Witte, "Topical Structure and Revision: An Exploratory Study", College Composition and Communication 34(3), 1983, pp. 313-341.
Susan Herring has been researching and writing about CMC since 1991. One of the first to employ linguistic methods of analysis to group interaction on the Internet, she has edited two collections (Computer-Mediated Communication: Linguistic, Social and Cross-Cultural Perspectives, John Benjamins, 1996; "Computer-Mediated Discourse Analysis," special issue of the Electronic Journal of Communication, 1996), and published articles on gender and CMC. Dr. Herring is currently Associate Professor of Linguistics at the University of Texas at Arlington.
Copyright © 1999 Institute of Electrical and Electronics Engineers. Reprinted from the CD ROM-based "Proceedings of the Thirty-Second Annual Hawaii International Conference on Systems Sciences," 1999 (January 5-8, 1999, Maui, Hawaii. This material is posted here with permission of the IEEE. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by sending a blank email message to firstname.lastname@example.org. By choosing to view this document, you agree to all provisions of the copyright laws protecting it.