Argumentation is one way of settling differences, and it is often prized by theorists as an alternative to violence and other less intellectual ways of managing conflicts (e.g., Perelman & Olbrechts-Tyteca 1969; Ehninger & Brockriede 1963). We academics have no difficulty at all in seeing that arguing is a dramatically different sort of thing than physical fighting. The clarity with which we see this, however, may not match the perceptions of our students, or the public at large. This paper explores the possibility of perceptual connections between arguing and violence among ordinary people.

1. Literature Review
At the 1997 Alta meeting, we reported some intriguing results about naive social actors’ understandings of argument (Benoit & Hample 1997). We had asked them to keep diaries about conflicts that they had avoided or cut short. But in reading the diary accounts, we frequently found ourselves wondering what could possibly have been avoided or cut off, because the narratives seemed very complete to us. After a number of re-readings, we decided that, unlike argumentation scholars, our respondents assume that there is a “violence slot” in the development of face to face arguments, and if nothing physically aggressive happened, the argument had not moved through all its potential phases. We also realized that they seemed not to count something as an argument at all if the central claim-and-disagreement were not explicit. That is, if no one had gotten around to announcing the disagreement, they thought the interaction was unfinished.
We followed that study up with a larger one, still using data from conflict diaries, but undertaking systematic coding of the accounts instead of a qualitative reading of them (Hample, Benoit, Houston, Purifoy, VanHyfte, & Wardwell 1998). We found that explicitness and destructiveness of the arguments in the diaries were correlated at r = .80, which is extraordinarily high. In other words, the more explicit (i.e., argument-like) a conflict was, the more destructive it was. For our respondents, arguments seem not to be alternatives to violence; instead, they appear to be companions to fights, or causes of them, or parts of them, or perhaps even essential to their nature.

That we were surprised by this is surely due more to our own perceptual blinders than anything else. Earlier work has shown that people are definitely fearful that their arguments will get out of hand, in spite of their earnestly cooperative intentions (Benoit 1982). Trapp (1990) documented a tendency for arguments to escalate into verbal aggression, and Infante, Chandler, and Rudd (1989) suggest that these out-of-control arguments may trigger spouse abuse. When naïve actors are asked to list the specific actions that may take place in a face to face argument, many of them specify one or more slots for threats and physical violence (Hample, Dean, Johnson, Kopp, & Ngoitz 1997). Gilbert (1997) explores the idea that both the standard practices and standard theories of argument are competitive, agonistic, and masculine. Somehow, though, our own commitments to argumentation as an alternative to violence, as a rational path to conflict management, as an interactive ideal in the face of conflicting wants, made us see all these negative indications about the nature of argument as being nothing but minor perceptual distortions, failures to understand the real nature of arguing. We continue to hold our original commitments, but we now feel a greater need to acknowledge, understand, and respect what non-specialists think about all this.

Let us begin by asking, What had to be so, in order for our diary studies to have come out the way they did? Recall that the key result, the one exercising us here, is that explicitness and destructiveness of arguments are very nearly synonymous for our informants. We understand explicitness to index the recognition that the encounter actually is an argumentative one. When the disagreement is unstated or the conflict merely implied, our diarists seemed to think that an argument hadn’t quite happened, that they were describing a conversation one could simply walk away from, without any argumentative commitments having been made. A clearly identifiable argument – an explicit one – was also a destructive one. Feelings would be hurt, bodies might be threatened, relationships could be put at risk.
Two explanations for this identity between explicitness and destructiveness occur to us. (1) This is the way the world is. People accurately and representatively reported their arguments to us. All, or virtually all, arguments really do implicate violence or its immediate possibility. (2) People have the prejudice that arguments are typically destructive. Therefore, whenever they identified something as being appropriate for a diary entry – perhaps as being a paradigm case of arguing – they had already judged that the episode had hurtful potential. Thus, destructiveness was a qualifying feature for something to be reported to us.
We have little to say here about the first possibility. We know that it is not so theoretically, because arguments can be productive and emotionally positive experiences. The equivalence of arguing and hurting may well be an accurate summary of some people’s personal experience with conflicts (Hample, in press), but this is not the case for everyone. The design of the study we report here controls for this first possibility by making it inoperative in our stimulus materials.
The second possibility is the one that mainly concerns us. On reflection, we believe that explanation (2) could have two different causes. The first is that people’s “destructiveness prejudice” is as deeply seated as some other deleterious stereotypes. Whenever they see an argument, they think they are also seeing danger of some sort. This might be a defensive reaction against the known possibility of hurt, or might reflect a negativity bias in perceiving others and their actions (see Coovert & Reeder 1990). The second possible cause is that people simply misunderstand what is meant by “argument” (or, less presumptuously, that naïve actor’s definitions are just different than ours), and that they can be educated toward a view that more closely resembles the academic one.

Our research strategy is a fairly simple one, at bottom. We created a series of argument vignettes that were each explicit or implicit, destructive or constructive. Then we asked respondents to rate these on scales that would reflect perceptions of explicitness and destructiveness. By providing these scenarios ourselves, and systematically varying the two key variables, we removed possibility (1) from consideration. Even if the world nearly always does generate explicit and damaging arguments, our design doesn’t.
We anticipate two possible sorts of outcomes, which correspond to the two possible reasons for explanation (2). If our respondents continue to see explicitness and destructiveness as highly correlated (even though they have been manipulated to be orthogonal in our design), we will be confident that we are observing a persistent perceptual bias. Anticipating the possibility of this outcome, we included tests of a trait connected to conflict perceptions, the tendency to take conflict personally (TCP) (Hample & Dallinger 1995). Should this perceptual effect appear, the TCP data might be a first step in explaining its etiology. If, on the other hand, the correlation between explicitness and destructiveness disappears, we will conclude that our diary respondents were simply working with a different definition of “argument,” one that will not be unusually difficult to change through standard instruction; that is, the earlier association was due to their category “argument,” rather than to their perceptions or biases about it. Our hope is that the correlation dissipates, but we have no grounds for predicting whether it will or not, so we offer no formal hypothesis.

2. Method
2.1. Respondents
Data were provided by 303 students enrolled in undergraduate communication classes at the first author’s institution. 53% were male, and the sample’s mean age was 20.3 years. The sample was distributed across the four undergraduate years, with 20-30% belonging to each class.

2.2 Procedure
Respondents were given a booklet. After giving consent, they answered some demographic questions and filled out the trait version of the Taking Conflict Personally scale (Hample & Dallinger 1995). The respondent then read a brief argument scenario, was instructed to imagine himself/herself participating in the interaction, and then filled out several scales about the imagined interaction. Then each respondent repeated these last procedures with a second scenario. After finishing, students were debriefed.

2.3. Scenarios and Instructions
We wrote 16 scenarios, which were randomly distributed throughout the sample. We based them, as far as we could, on actual reports from the diary studies. The scenarios were labeled “possible interactions” for respondents. These vignettes were designed to represent several conditions. Half were explicit, and half implicit; half were constructive, and half destructive; half involved an argument with a roommate, and half with a romantic partner. Each condition was represented twice in the collection of scenarios, yielding a 2x2x2x2 (explicit x constructive x relationship x replication) design.
To ease the understanding of our results, please notice that “replication” will always refer here to a different example of, say, an explicit, constructive, roommate argument. This is easily confused with the fact that each respondent read two scenarios; these will be called scenario (or vignette) 1 and 2. The “replications” have nothing to do with the scenario series, so that the second explicit/constructive/roommate argument was read first just as often as it was read second.
Each scenario was preceded by these instructions: “Please read the brief description of a possible interaction you might have. Pause for a moment to visualize it, and to imagine you’re actually in it. How would you react? What would you do or say? How would you feel? After you’ve taken a moment to do that, please respond to the items we’ve provided, according to how they apply to the interaction we’ve supplied.” Here is an example of one of the scenarios, instantiating a destructive, not-explicit argument with a romantic partner: “You have a long distance relationship. You haven’t seen your romantic partner for a month and when s/he comes to visit, you go out with some of your friends. When you are out dancing, your romantic partner trips you, as a sort of bad joke.” Following the scenario, 16 items (described below) were provided, along with standard instructions on how to respond.

2.4. Instruments
Taking Conflict Personally (TCP) was measured by means of a 37 item Likert instrument designed to produce scores on six subscales (Hample & Dallinger 1995). Cronbach’s alphas for the subscales are as follows: direct personalization, .82 (omitting item 1); persecution feelings, .74; stress reactions, .65 (omitting item 27); positive relational effects, .74; negative relational effects, .79; and like/dislike valence, .75 (omitting item 20).

Constructiveness and explicitness were measured by eight semantic differential items each. These items were generated for this study. They are as follows (where R indicates reverse scoring, and item numbers are shown).
Constructiveness:
– constructive/destructive [1],
– good/bad [2],
– harmful/beneficial [4R],
– helpful/damaging [8],
– a controlled interaction/not a controlled interaction [10],
– would harm our relationship/would not harm our relationship [12R],
– violent/nonviolent [13R],
– positive/negative [15].
Explicitness:
– explicitly an argument/not explicitly an argument [3],
– hid my feelings/ expressed my feelings [5R],
– a conflict/not a conflict [6],
– didn’t give reasons for what I said or did/did give my reasons for what I said or did [7R],
– disagreement between us not apparent/disagreement between us was apparent [9R],
– I communicated clearly/I didn’t communicate clearly [11],
– said what I thought/did not say what I thought [14], and
– everything would be out in the open/everything would not be out in the open [16].

We ran a series of exploratory factor analyses on these scales. Since each respondent filled out the scales for two scenarios, we conducted separate factor analyses for each scenario series. The destructiveness items performed more or less as expected, loading together for both scenario series, except that item 15 had to be dropped. On analysis, the explicitness items proved to be measuring two different things. The first, which we will continue to call explicitness, consists of items 3, 6, and 9. The other, which in hindsight appears to be assessing the presence or absence of full disclosure, we will call disclosiveness. We therefore formed scales, as follows.

The constructive/destructive scale consists of items 1, 2, 4, 8, 10, 12, and 13, and produces Cronbach’s alphas of .89 and .90 for the two scenarios. The explicitness measure includes items 3, 6, and 9, with alphas of .71 and .76. The disclosiveness scale consists of items 5, 7, 11, 14, and 16, and yields alphas of .84 and .87. The directions of scoring were such that a high score on the first variable shows that respondents felt the conflict would be destructive, a high score on the second means that respondents thought the conflict would be implicit, and a high score on the third indicates nondisclosiveness.

3. Results
3.1. The Associations Among Destructiveness, Explicitness, and Disclosiveness
Perceived destructiveness and explicitness were correlated with one another. The first scenario series produces r = -.48 (p<.001), and the second r = -.51 (p<.001). These results replicate both Benoit and Hample’s (1997) and Hample, Benoit, et al.’s (1998) findings, because both these papers report an association between destructiveness and explicitness, as in the present report. Perceived destructiveness and nondisclosiveness were also correlated with one another. In the first scenario, r = .35 (p<.001), and in the second, r = .24 (p< .001). Finally, the associations between disclosiveness and explicitness are r = .06 (ns) for the first series of vignettes, and r = .33 (p<.001) for the second.

The two previous studies examined respondents’ argument diaries, and the researchers observed a strong tendency for destructiveness and explicitness to co-occur in those accounts. The systematic manipulation of destructiveness and explicitness in the present design guards against such co-occurence, but therefore produced a systematically different sample of arguments to be rated. Consequently, we also examined the destructiveness-explicitness correlations in the present study after dividing the data set by manipulated destructiveness and explicitness. Table 1 displays the correlations between perceived destructiveness and implicitness, by manipulation condition.

Table 1 Correlations between destructiveness and implicitness, for each combination of manipulated destructiveness and explicitness, for both scenario series.

The overall destructive/explicit correlations of -.48 and -.51 reported above are obviously depressed by the constructive/implicit conditions, which would probably have been rarest in the diaries data. The destructive/explicit conditions are perhaps most representative of the diaries data. This manipulation produces a moderate association between perceived destructiveness and perceived explicitness. The overall results, combined with the correlations subdivided by manipulation condition, indicate that our respondents persist in associating explicitness and destructiveness, but not at the exceptionally high levels reported for diarists (Hample, Benoit, et al. 1988). When these two variables are manipulated to be orthogonal in the stimulus set, as was done here, the destructiveness/explicitness correlation is somewhat reduced, but still clearly evident.
These results are supportive of explanation 2, because the design temporarily eliminates the possibility that the world supplies only explicit, damaging arguments to people.

Taking Conflict Personally
The TCP scales were correlated with respondents’ ratings of the destructiveness, implicitness, and nondisclosiveness of each scenario.

Since each person responded to two vignettes, each pair of variables produces two correlations. Results appear in Table 2. The table shows few significant results, and little consistency among them. Since the scenarios were randomly distributed, and each occurred as often as the first stimulus as it did the second, the differences between the pairs of columns can only be attributed to some sort of fatigue effect. The table gives no evidence that there is any connection between people’s trait TCP and their estimate that the argument would have been destructive or explicit, and weak evidence of a small connection between TCP and estimates of  disclosiveness (such that people high in TCP may be somewhat less likely to be disclosive). In short, TCP appears to have little effect on one’s perception of an argument’s destructiveness, explicitness, or disclosiveness.

Table 2. Correlations between TCP variables and the rated destructiveness, implicitness, and disclosiveness of vignettes from scenario series 1 and 2.

3.3. Perceived Destructiveness, Explicitness, and Disclosiveness as Dependent Variables
This set of results clears the way for an examination of whether the manipulations (destructiveness, explicitness, relationship with other, and replication) had effects on perceived destructiveness, explicitness, or disclosiveness. We therefore undertook 2x2x2x2 ANOVAs, doing each analysis twice, once for the respondents’ first vignette, and once for the second.
When perceived destructiveness is the dependent variable, the first vignette yields several significant results. Significant main effects appear for the constructiveness manipulation (F = 111.4, df = 1, 282, p .001; the destructively intended scenarios have higher means, as expected, 22.7 versus 16.5), the explicitness manipulation (F = 5.4, df = 1, 282, p .05; the implicit arguments are rated as more destructive, 20.5 versus 18.8), relationship with other (F = 19.4, df = 1, 282, p .001; the roommate conflicts are more destructive than those with romantic partners, 21.0 versus 18.6), and replication (F = 4.4, df = 1, 282, p .05). The only significant interactions involved the replication manipulation, and are therefore of no substantive interest here.
The analysis was repeated for the respondents’ second imagined interaction, with similar results. Significant main effects appear for the constructiveness manipulation (F = 122.0, df = 1, 281, p .001; the destructively intended vignettes are seen as more destructive, 24.4 versus 17.2), the explicitness manipulation (F = 24.1, df = 1, 281, p .001; the implicit interactions are again more destructive, 22.8 versus 19.3), and the relationship with other (F = 11.3, df = 1, 281, p .001; the roommate conflicts are again seen as more destructive, 22.3 versus 19.8), but not for replication (F 1). The only significant interaction not involving the replication factor is a two-way interaction between manipulated explicitness and relationship with other (F = 5.1, df = 1, 281, p .05). The means, displayed in Table 3, indicate that the effects of explicitness are strongly influenced by relationship: the destructiveness of the inter-action is greatest when romantic partners have implicit conflicts, and is most manageable when romantic partners have explicit arguments.

Table 3. Destructiveness means for the second vignette, by explicitness and target

The destructiveness results are nicely consistent. The main effects for constructiveness indicate merely that the manipulation worked. The main effects indicate that arguments with one’s roommate (compared with one’s romantic partner) are felt to be more dangerous, as are implicit conflicts. This last finding is inconsistent with the positive association between perceived destructiveness and perceived explicitness, and raises the question of whether the respondents and experimenters are viewing explicitness in the same way.

Similar analyses of variance were done, using explicitness as the dependent variable. Results for the first scenario series are as follows. Significant main effects appear for replication (F = 7.0, df = 1, 284, p 01) and destructiveness (F = 50.2, df = 1, 284, p .001, with means indicating that the destructively intended vignettes are seen as more explicit, 7.2 versus 9.4). Argument partner has no effect (F = 1.0, df = 1, 284, ns), and neither does manipulated explicitness (F 1). This last results points to a manipulation failure. The only significant interaction not involving the replication factor is a two-way between argument partner and destructiveness (F = 6.1, df = 1, 284, p .05; means indicate that destructiveness had the least effect on romantic partners, with their explicitness means showing less difference than those of roommates).
The second scenario produced comparable results. Significant main effects appear again for replication (F = 4.8, df = 1, 281, p .05) and destructiveness (F = 122.1, df = 1, 281, p .001, with means again showing that the destructively intended vignettes are seen as being more explicit, 6.5 versus 9.9). The second scenario series also yields a significant effect for argument target (F = 15.9, df = 1, 281, p .001, with means indicating that the roommate conflicts are seen as more explicit, 7.6 versus 8.8). Here, too, however, we obtain the disappointing failure to confirm the manipulation, with the explicitness effect being insignificant (F = 1.9, df = 1, 281, ns, although the means are in the correct order, 8.0 versus 8.4). The only significant interaction effect not involving the replication factor is between explicitness and destructiveness (F = 4.4, df = 1, 281, p .05; means show that the greater perceived explicitness for destructive arguments is more marked for explicitly intended episodes).
Finally, we undertook parallel analyses, using perceived disclosiveness as the dependent variable. For the respondents’ first scenario, the only significant main effects are for relationship with other (F = 8.0, df = 1, 284, p .01; romantic partner arguments are seen as more disclosive, 12.4 versus 11.0) and replication (F = 7.2, df = 1, 284, p .01). The main effect for explicitness was not significant (F = 2.01, df = 1, 284, p = .15), although the means are the direction of explicit arguments having fuller disclosure, 11.2 versus 12.1. Several significant interactions not involving the replication factor appear. A two-way interaction between manipulated constructiveness and manipulated explicitness is significant (F = 4.2, df = 1, 284, p .05), as is a three-way interaction involving constructiveness, explicitness, and relationship with other (F = 7.3, df = 1, 284, p .01).

Table 4. Disclosiveness means for the first vignette, by explicitness, constructiveness, and target manipulations

The means illustrating this three-way interaction (and, of course, including the information necessary to see the two-way interaction), are in Table 4. These means indicate that the most disclosive arguments are those with romantic partners. For roommates, the arguments are seen as disclosive when they were intended to be explicit and constructive, and are seen as nondisclosive when they were intended to be implicit and constructive.

The parallel analysis on the second vignette for each respondent produced these results. Significant main effects appear for constructiveness (F = 11.1, df = 1, 282, p .001; destructive arguments are more disclosive, 11.2 versus 13.3), explicitness (F = 36.4, df = 1, 282, p .001; explicitly intended scenarios are seen as more disclosive, 10.4 versus 14.0), and replication (F = 14.1, df = 1, 282, p .001). The only significant interactions not involving the replication factor are between the constructive and explicitness manipulations (F = 12.6, df = 1, 282, p .001), and between the explicitness manipulation and relationship with other (F = 13.4, df = 1, 282, p .001).

The means for these interactions are in Table 5. These means indicate that the most disclosively perceived arguments are those that were supposed to be implicit and constructive for romantic partners. The least disclosively perceived ones are those that were intended to be explicit and between romantic partners. The results for romantic partners are the most variable, suggesting that these relationships involve more exaggeration of (or, perhaps, more salience for) the manipulations.

Table 5. Disclosiveness means for the second vignette, by explicitness, constructiveness, and target manipulations

The results for perception of disclosiveness are not entirely consistent. The two scenarios do not produce quite the same pattern of significant results, in spite of their containing the same scenarios in the same proportions. Results suggest that arguments with romantic partners are seen as having fuller disclosure, and that destructive arguments are also seen the same way, although these effects are qualified by interactions. When compared to the explicitness results, these findings suggest that we may have been somewhat more successful in manipulating disclosiveness than explicitness, in spite of having intended only to vary the latter.

4. Discussion
In this final section, we wish to discuss our leading results, and then to return to the issues that stimulated the study.
Perhaps the two most interesting empirical findings in this paper are that (1) destructiveness and explicitness are positively correlated, and (2) roommate conflicts are more destructive than conflicts with romantic partners. Both findings deserve some comment.

First, let us consider the destructive-explicit association. As in the diary studies, our respondents made a clear connection between the explicitness of an argument and its inherent danger. The diary data (Hample, Benoit, et al. 1998) produced a stronger correlation than appeared here, and the effectiveness of this study’s explicitness manipulation is open to question. Still, this investigation both replicates and triangulates the earlier finding. The chief benefit of the present design is that our attempt to vary destructiveness and explicitness orthogonally means that the earlier findings cannot be explained by our instructions to diarists, or by their search for paradigm cases of face to face argument.
We were faced with an unanticipated complication in our study of the association between destructiveness and explicitness, however, and this centered around our own understanding of explicitness. We generated a number of scales to reflect our own construct for explicitness: that an episode would have clear disagreement, that it would obviously be mutually framed as an argument, that both people would speak their minds, and that they would put their claims and counter-claims on the table. Our respondents had a more sophisticated view of all this, however, and saw two things being indexed in those scales: the explicitness and obviousness of the episode-as-argument, and the degree to which our respondents would have been willing to express their true thoughts and feelings. Dividing this into two separate scales was not a problem, of course. The problem occurred when we tried to translate our understanding of explicitness into varying scenarios. The explicitness manipulation did pretty much work as far as disclosiveness was concerned, but it did not produce supportive means on the measure of perceived explicitness. This necessarily qualifies all our findings relating to the so-called manipulation of explicitness.
On reflection, we now think that it may be important to consider that some sorts of argument may involve a failure to self-disclose, an unwillingness to expose one’s thoughts and feelings to the other. This might be thought dangerous and destructive in and of itself, and might also be seen as symptomatic of an already damaged interaction or relationship. Thus, arguments in which participants do not say what they think are perceived as bad, harmful, and damaging, but this may have less to do with recognizing whether the episode is an argument, than with the view that a non-disclosive interaction is already fundamentally flawed. So a nondisclosive argument might indicate damage, rather than cause it.

The second empirical finding of note is that roommate conflicts were rated as more fraught with danger than romantic conflicts. One would suppose that the stakes would be higher with romantic partners, making intense hurt and extreme danger more possible in that setting. But roommates evoked more apprehension here. We wonder if two factors – other than the stakes – might be at work here. First, in our sample, people were living with their roommates, but not necessarily with their romantic partners. One can simply go home after a date, but still has to sleep in the apartment. This would make conflicts less avoidable for roommates, and so more dangerous. Second, the intimacy of a romantic relationship may well have generated norms for handling conflict, in part because of the greater value placed on the relationship by both parties. Table 3 showed an interaction effect, such that implicit arguments were rated as particularly destructive for romantic couples. Even though intimate partners may have better developed conflict management norms, ambiguous episodes can be threatening, and the high stakes may exacerbate this more than for roommates.

Perhaps the most substantial issue raised by our results is whether they are compatible with those of Benoit and Hample (1997) and Hample, Benoit, et al. (1998). The present findings replicate the earlier ones. The connection between destructive potential and explicitness was unmistakable in all these investigations. In the present study, we created a stimulus set that destroyed any possible natural connection between explicitness and destructiveness. This design strategy permitted us to test whether the earlier result was due to the peculiarity of the argument sample we were dealing with. In fact, the clear association between perceived destructiveness and perceived explicitness does not appear to depend on the argument sample, and this considerably improves our confidence in the finding. However, it also points toward the more troubling explanation for the association, namely, that people have a fundamental, stereotyped pessimism about arguments.
This leads us toward several conclusions regarding our title question, Must arguments be explicit and violent? (1) Left to their own devices, and asked to identify clear arguments from their lives, people will report explicit, dangerous episodes. (2) Given a systematically neutral stimulus set, people still see a strong connection between explicitness and destructiveness. (3) This derives from their perception of what an argument is, and this view is one argumentation scholars wish to alter. (4) Therefore, we should worry about our students’ most basic understandings of argumentation, and not take for granted that they believe us when we tell them that arguments are alternatives to violence.

REFERENCES
Benoit, P. J., & Hample, D. (1997, July). The meaning of two cultural categories: Avoiding interpersonal arguments or cutting them short. Paper presented at the Summer Conference on Argument, Alta, UT.
Benoit, P. J. (1982, November). The naive social actor’s concept of argument. Paper presented at the Speech Communication Association, Louisville, KY.
Coovert, M. D., & Reeder, G. D. (1990). Negativity effects in impression formation: The role of unit formation and schematic expectations. Journal of Experimental Social Psychology 26, 49-62.
Ehninger, D., & Brockriede, W. (1963). Decision by Debate. New York: Dodd.
Gilbert, M. A. (1997). Coalescent Argumentation. Mahwah, NJ: Erlbaum.
Hample, D. (in press). The life space of personalized conflicts. Communication Yearbook.
Hample, D., & Dallinger, J.M. (1995). A Lewinian perspective on Taking Conflict Personally: Revision, refinement, and validation of the instrument. Communication Quarterly 43, 297-319.
Hample, D., Dean, C., Johnson, A., Kopp, L., & Ngoitz, A. (1997, May). Conflict as a MOP in conversational behavior. Paper presented to the annual meeting of the International Communication Association, Montreal.
Infante, D. A., Chandler, T. A., & Rudd, J. E. (1989). Test of an argumentative skill deficiency model of interspousal violence. Communication Monographs 56, 163-177.
Perelman, C., & Olbrechts-Tyteca, L. (1969). The New Rhetoric: A Treatise on Argument (trans. J. Wilkinson & P. Weaver). Notre Dame, IN: University of Notre Dame Press.
Trapp, R. (1990). Arguments in interpersonal relationships. In: R. Trapp & J. Schuetz (Eds.), Perspectives on Argumentation: Essays in Honor of Wayne Brockriede (pp. 43-54). Prospect Heights, IL: Waveland.

1. Practical Arguments
Our point of departure is the practical syllogism. The invention is Aristotle’s and the interpretation we give it is Anscombe’s (Anscombe, 1957). As is well-known, the standard syllogism is a discursive entity, an n-tuple of declarative sentences, of which the terminal member is the conclusion and the rest are premisses. In contrast, a practical syllogism is a mixed structure, part discursive and part non-discursive. The difference shows up in the conclusions of the two structures. In a standard syllogism, the conclusion is a sentence; in a practical syllogism the conclusion is an action. It is useful to compare practical syllogisms with deontic or prudential arguments. A simple example of such is:
1. If you are late home from the movies, you’ll irritate and worry your mother
2. So, you shouldn’t be late.

It is easy to construct what we could call the practical syllogisation of this argument. It is the ordered pair in which the first member is the premiss of the deontic argument

1* If you are late home from the movies, you’ll irritate and worry your mother.

and in which the second member, the conclusion, is not the sentence which bids the addressee not to be late, but is simply the addressee’s not being late.Thus the conclusion of our practical syllogism is the action advocated by the conclusion of the preceding deontic argument.

The distinction between deontic-prudential arguments and practical syllogisms calls to mind the old maxim that talk is cheap. `Cheap’ in turn suggests `suboptimal’, and, in some respects, this is precisely what can be claimed by deontic-prudential arguments in contrast with their practical syllogisations. It is one thing to get an addressee to concede that he should do such-and-such; it is another, and often better thing, that he actually do it. Better the cheque in the mail than `The cheque is in the mail’. We may say in a quite general way that a practical syllogism is the consummation of a deontic arguer’s intent.

In this note we propose to expand the concept of practical syllogism in a slight but natural way. We shall attempt to show that modest though the extension might be, it produces results of genuine consequence for the theory of argument. In our proposal, a second way of being a practical syllogism is one in which one or more of the premisses is an action rather than a sentence. It is a point worth emphasising that the conclusions and, as we now may say, the premisses that make for practical syllogisms are role-specific. Any action by any agent at any time, make for a true proposition, namely the proposition ascribing that action to that agent at that time. Any of these truths is available in principle as the conclusion or as a premiss of some or other bit of argument that may chance at a time to bubble out of the dialectical soup of the human community. Such arguments are not made into practical syllogisms in consequence of this fact; for it is the actions themselves, not the sentences they make true, that are the irreducible components of practical syllogisms. In Aristotle’s conception of it, the action that is the conclusion must be the action of the party to whom the argument is addressed. In our extension of it, the action that is the premiss of a practical syllogism must be the action of the maker of the argument, not his addressee. So there is an agent-specific asymmetry between, as we shall now say, conclusionally practical syllogisms and premissorily practical syllogisms.

There is a further asymmetry. Let <‘P’, A> be a conclusionally practical syllogism. Let ‘A’ be the sentence in which this action A is attributed to its agent. Then, in general, a standard syllogism <‘P’, ‘A’> is not preservable from the practical syllogism <‘P’, A> under replacement of A by ‘A’.[i] On the other hand, consider a simple case of a premissorily practical syllogism. Suppose that Joe will be elected Treasurer if and only if Henry, Sarah, Frank and John vote yes. Imagine that John is attempting to construct an argument whose conclusion is that Joe is elected. Joe adds the following true premisses.

2. ‘Henry has voted yes’
3. ‘Frank has voted yes’
4. ‘Sarah has voted yes’.

The desired conclusion that Joe will be elected Treasurer requires a further premiss. So:

5. [John simply votes yes.]

In this, our two asymmetries are evident. For one thing, the action which serves as the clinching premiss must be John’s, the speaker of the argument, rather than Sarah’s or Henry’s. But, secondly, if our previously practical syllogism is correct, there is a correct standard syllogism got by replacement of the action that constitutes premiss (5) of the former with the sentence `John votes yes’, which correctly attributes that action to John.

Essential to both types of practical syllogism, and corresponding to the parameter of role specificity, is the element of participant control. If my conclusionally practical syllogism that you do so-and-so is good, then that it is so lies essentially in your power, not mine. All that rests with me is to show that you should do so-and-so. But if my intent is to produce a practical syllogism rather than a deontic-prudential argument, the premisses are up to me to select and present; but the conclusion finally is up to you. This other-party dependency is missing in the case of premissorily practical syllogisms. To recur to our example, the argument cannot succeed without premise (5), and yet premise (5) in an action entirely up to John, the person whose argument it is. Similarly, the corresponding standard syllogism has no chance without the sententialization of premiss (5), i.e. ‘John voted yes’. But that premiss is true if and only if John voted yes; which, again, is entirely up to John.

2. Ad baculum reasoning
It is perhaps not surprising that fallacy theorists and argument analysts should have been preoccupied with the idea that there is something inherently defective about ad baculum arguments. Our own view is that those comparatively few writers are correct who, like Walton [1992] and Woods [1987], [1995], see the ad baculum as a form of prudential argument which, when bad, cannot have been made bad simply because it pivoted on the factor of threat. We lack the space to expatiate on this prudential perspective, promising as we think it is.[ii] Instead we shall take the Woods–Walton approach a step further. We shall show that
1. ad baculum arguments are systematically connected to premissorily practical syllogisms;
2. they are in a sense to be explained always a more benign and welcome form of argument than their counterpart practical syllogisms;
3. in vindication of something theorists such as Walton have been  saying – perhaps with insufficient explicit motivation – arguments from negative consequences are not as such ad baculum arguments (that is, ad baculum arguments are a proper subset of negative consequence arguments); and
4. (recurring to point (2)), although some theorists have been aware of the importance of utility functions in the analysis of ad baculum arguments, there are always utility-functional considerations which favour recourse to ad baculum argument over their counterpart practical syllogisms.

Let us now see how it is that our analysis of ad baculum arguments give rise to these four consequences. We consider in turn three arguments of a type well-known in the recent literature. They are:
1. collective bargaining arguments;
2. the mugger’s argument and
3. anti-smoking arguments.

2.1 Collective bargaining
For expository convenience we consider a simplified case. We assume that in the present example both parties, workers and management alike, are satisfied that a threat to strike is sincere and that a strike would encumber management with higher costs than would a settlement in the near vicinity of the union’s most recent offer. Even so, consider the following action-matrix, an ordered 2-tuple.
1. The workers strike (S)
2. The management yields (Y)

Schematically our action-matrix is
1* S
2* Y

‘S’ and ‘Y’ are abbreviations of (1) and (2), which in turn report certain action-facts. In the circumstances of the case, the episode characterised by <S, Y> is costlier to each party than an available alternative. In real-life situations, this is not always the case, of course, and in any event, calculating the actual cost-benefit spread over actual option spaces can be a fairly complex matter.  Even so, we know that one of the alternatives is the one we now describe; and we also know that in general it appears to yield a better cost-benefit payoff for both parties. We represent this option as a dialogue between the workers’ representative W and management’s spokesman M. As before, ‘S’ denotes the strike-action and ‘Y’ management’s action of yielding to the present demands of the workers.
W: ‘If ¬ Y, then S’
M: Y.

As we see, W makes an explicit threat. It is a conditionalisation of S on the negation of Y in our action-matrix <S, Y>. M’s response to W is an action, a capitulation to W’s demand. If we could think of the sequence <‘If ¬ Y, then S’Y> as an argument, then not only is it a cross-agent argument; it is a conclusionally practical syllogism. Its most distinctive feature, however, is that it is a substitute for a premissorily practical syllogism, which is what our actionmatrix <S, Y> in effect is. In the W-M dialogue (or quasi-dialogue), an action which is in the control of W to perform and which, if performed, would serve as a premiss in the practical syllogism <S, Y>, is only threatened. In our simplified example, the threat is justified on simple cost-benefit grounds. It is less costly to threaten to strike than to strike, and it is no more costly to yield to the threat of a strike than to a strike.
We propose that dialogues or quasi-dialogues of the W-M type are prototypes of ad baculum argumentation. If so, it is easy to see the systematic link between ad baculum arguments and practical syllogisms. The threat that constitutes the dialogue as an ad baculum threatens an action which is within the threatener’s power to effect, and which if effected would produce a premiss in the practical syllogism <S, …>, where … holds a place for, but does not guarantee, the appearance of the intended M-action Y.
It is also apparent that ad baculum arguments have clear advantages over the premissorily practical syllogisms, to which they are systematically linked. Here is a case in which `Talk is cheap’ is a virtue. The threat to strike possesses at least the following advantages over striking. Even an efficacious threat to strike is in general, as we have seen, a less costly inducement to yield than yielding to an actual strike. Moreover, talking about striking, rather than striking, provides the contesting partners with a larger deliberation-space than simply striking. Thus ad baculum contentions are dialectically more efficient (to say the least) than the premissorily practical syllogisms to which they are linked. We take it, then, that the characteristic features (1) and (2), cited above, may now be claimed for ad baculum exchanges. These same features will be apparent in our next example, the mugger.

2.2 The mugger
Here too, there are two parties, M, the mugger, and V, the victim, and an action-matrix <K, T> in which K is the killing of V by M and T is M’s getting V’s money. As before <K, T> can be likened to a premissorily practical syllogism, and as before it is a less good thing than its counterpart ad baculum, in which the action-premiss K is replaced by a discursive premiss which threatens K. It is bad enough to be threatened with death, but for most people in most circumstances it is a better thing than death itself. The mugger’s ad baculum achieves two things at once. It identifies a situation in the joint option space which itself is constituted by the premissorily practical syllogism <K, T>. And it gives the addressee the option of replacing the muggers’ practical syllogism with his own cross-agent conclusionally practical syllogism

M: ‘If ¬ T, then K’
V: T.

As before, the conclusionally practical syllogisms confers on M all the benefits conferred by the premissorily practical syllogism, yet sparing V the extreme cost of that option. Either way, V loses his money. But in only one of these ways does he lose his life.

2.3 Anti-smoking arguments
Again we simplify. We shall take it that in some non-trivial sense, habitual cigarette-smoking shortens a smoker’s lifespan. If so, then we could expect to find instances of the matrix

Sm
D

(‘Sm’ for ‘The subject was an habitual smoker’ and ‘D’ for ‘The subject died earlier than would have been the case otherwise’). Is there an ad baculum counterpart of this sequence <Sm, D>? If so, it would be something like:

P1: ‘If ¬ D then Sm’
P2: D

(where P1 and P2 are respectively the ad baculum-maker and his addressee.) We see that the absurdity of this reconstruction as self-announcing. This is tantamount to a proof that negative-consequence arguments are not just as they stand ad baculum arguments. If this is right, the rejection of the present example by the ad baculum model will show up in structural features of the model. If the anti-smoking argument were an ad baculum, then the sequence

Sm
D

would be construable as a premissorily practical syllogism. For this to be so, two conditions require fulfilment. One is that Sm be an action-premiss, and the other is that Sm be the action of the argument-maker. But as the example shows, this is not the case. Similar difficulties, and then some, apply to the interpretation of the would-be ad baculum

P1: ‘If ¬ D then Sm’
P2: D

The reader will note that we have conformed the present example to the ad baculum structure recognised in our model. It is significant that it gives rise to such nonsense. For one thing, it is hard to conceive of ‘If you don’t die, then you are an habitual smoker’ as any kind of threat (In fact, it may be wondered what are the truth conditions of this fabulous conditional.) For another, D can hardly be represented as P2’s action-conclusion, since in no direct way is his death in his own control. Thus the sequence <‘If ¬ D, then Sm’, D> is not representable as a conclusionally practical syllogism. We see, then, that if we opt for an analysis of the ad baculum in which premissorily and conclusionally practical syllogisms play a load-bearing role, the anti-smoking argument cannot be made out to be ad baculum. Central to this result is the fact that even if Nature herself threatens a certain fate for the smoker and even if I know this, I cannot threaten the same thing on Nature’s behalf, so to speak.

The same holds of Pascal’s Wager, in which the Wagerer cites God’s threat to the Christian sceptic. But citing a threat is not making a threat. Once cannot issue God’s threats, except that one is God, anymore than one can catch Yogi Berra’s catches except where one is Yogi Berra. We conclude, therefore, that contrary to recent speculation to the contrary (Woods,1987 and 1995), Pascal’s Wager is not an ad baculum for Pascal, though it would be for God. This being so, Walton is right to say, in effect, that negative-consequence arguments and ad baculum arguments share no more intimacy than a set-theoretic intersection which chances to be a proper subset of each.

We said that utility functions play an important role in the analysis of practical argument. At one level, there is a constant utility-functional component. The arguer seeks to give the addressee the option of conceding, hence of avoiding the cost of looking stupid (not to put too fine a point on it). In other respects, utility-functions bite more differentially. When the mugger makes an intervention ad baculum, he is predicting his victim’s deployment of utility functions in a context of menace imposed by the mugger himself. But, as we have already suggested, there is a further respect in which ad baculum arguments are the result of utility functions of both arguer and addressee alike. For the arguer to forward an ad baculum is a reflection of a cost-benefit analysis which induces the arguer to favour the ad baculum over its counterpart premissorily practical syllogism. On the other hand, the ad baculum-maker also anticipates a favourable cost-benefit determination by his addressee in which it is obviously preferable to yield a benefit under threat of death than to suffer the loss of the benefit as a result of one’s death.

The above discussion shows that utility, actions and time play a central part in our understanding of the ad baculum. The underlying logical model is propositional logic enriched by temporal flow and action symbols. The next section develops such a model in some generality. Such models can be applied to other areas such as the analysis of natural language conditionals, but we shall leave that for another time. Because space is limited we shall not dwell in detail on the way the analysis of Sections 1 and 2 is represented in the model. It will, in any case, be obvious to the interested reader. We shall also give a fairly realistic example.

3. Description of a Basic Model
We imagine that we are moving through a flow of time. Time is discrete (day after day?) and moves by the performance of actions. So if we are at time t we can move to time t+1 by performing some action a.

3.1 Example
A and B are arguing about something. The database contains

1. p ^ q -> r
2. p

B is desperate to deduce r. He controls an action whose postcondition is q. By performing this action he is practically inserting q into the database and thus enabling the deduction of r. In the language of Section 1, B is in the process of constructing a premissorily practical syllogism.

3.2 Example
Jobless John (JJ) has an old car which he insured with the Universal Insurance Company. According to the terms of the insurance his coverage expires December 31st 1990. As is common practice with many insurance companies, if JJ pays his premium by 31.1.91, his insurance coverage is renewed from 1.1.91 to 31.12.91. So for example, if JJ forgets to pay on 1.1.91 and has an accident on 15.1.91, he can still pay his premium on 20.1.91 and be covered on the 15.1.91 accident.
In our story, JJ has no money and does not pay his premium. On 15.1.91 he bumps into Richy Rich’s (RR) Rolls Royce, causing extensive damage. RR now has a problem. It is clear that JJ cannot pay his premium. If he doesn’t, then he is not covered, and RR cannot collect from JJ’s company. RR cannot of course collect from JJ. On the other hand, RR collects from his own insurance company, he will lose his 56% no-claims bonus. Let us give some utility values.

JJ’s premium is $500
RR loss of the no-claims bonus is worth $3000
RR damage is assessed at $8000.

It is clearly worthwhile for RR to pay JJ’s premium provided JJ is co-operative. We assume the factual circumstances of the accident, D accident, strongly support of q = ‘J is at fault’. This means that the database can probably prove that JJ is at fault even if JJ denies fault. However, it is much simpler if q1= ‘JJ admits fault’ is available. Let us now construct the story formally.

3.3 Propositions

p = JJ damages RR’s car
q = it is JJ’s fault
q1 = JJ formally admits fault.
c = JJ is covered
r = JJ insurance pays
Daccident = facts about accident

3.4   Actions    

RR wants r to follow. For that he needs to generate q and c. The simplest course of action is for JJ to make q1 true at t+1 and give a commitment to perform a, then get $750 from RR, then go ahead and perform a; and then RR can claim from the insurance company. JJ can threaten RR that he will not renew his insurance unless he receives $750 from RR. This is a legitimate threat. Although JJ is at fault and may face a damages claim from RR, it is clear that simply by not renewing his insurance, he creates a premissorily practical syllogism of particular consequence for RR. This is why, in effect, RR is trying to persuade JJ to produce a different premissorily practical syllogism, in which the practical premiss of not renewing is replaced by its action-negation.

We can now summarise what an ad baculum fallacy is. It is not a fallacy. At worst it is an incompetent threatening move, which is either illegal (since preconditions do not hold) or ineffective (since it has a low utility threat for the postcondition).

NOTES
i. That this is so is indicated by the fact that the more natural candidate for the standard syllogistic counterpart of the present practical syllogism would be the deontic argument whose conclusion is, ‘You ought to do A’.
ii. But see Wreen [1995].

REFERENCES
Anscombe 1957. G. E. M. Anscombe, Intention, Ithaca: Cornell Univeristy Press.
Walton 1992. Douglas Walton, The Place of Emotion in Argument, University Park: The Pennsylvania State University Press.
Woods 1987. John Woods, “Ad Baculum, Self-interest and Pascal’s Wager”, In: Frans H. van Eemeren et al. (eds.), Argumentation: Across the Lines of Discipline, Dordrecht: Foris, 343–349.
Woods 1995. John Woods, “Appeal to Force”, In: Hans V. Hansen and Robert C. Pinto (eds.), Fallacies: Classical and Contemporary Readings, University Park: The Pennsylvania State University Press, 240–250.
Wreen 1995. Michael Wreen, “Knockdown Arguments”, Informal Logic 17, 316–336.

To the memory of Theodoor Jan Krabbe (1941-1996) [i]

1. Introduction: The Self-Gratulatory Argument
There can be no doubt that this is a perfect morning for the study of quasi-logical arguments. Otherwise, to say it bluntly, our hosts wouldn’t have put it on the program. Or would it be quasi-logical to say so? Anyhow, quasi-logical arguments are what’s up, and I’m much honored that you have all come to join me in this enterprise.
Of lectures on the quasi-logical there are exactly two types, either they are long or else they are short. Fortunately, I hate long lectures. This is fortunate for you, but also for me. Why may I rejoice in my own abhorrence of lengthy lectures? Well, that can be argued thus: suppose I liked long lectures, then I would certainly give one right now and be bound to hearing it out; but to hear my own long lecture would be a bad thing, since I happen to hate long lectures. So am I ever happy to hate long lectures!
That was an argument. Was that a quasi-logical argument? Yes, it was. It was meant to give you a taste of the quasi-logical, that is, to put it briefly, of a style of reasoning that unwarrantedly takes on the trappings of logical or mathematical rigor.[ii]

So, if this was a quasi-logical argument, what is its dialectic? Now wait. Give me a break. We’ll get to that later. It is certainly my intention in this lecture to show some of the dialectic of this argument; that is to expound in a profile of dialogue some of the moves and countermoves available to its discussants. But first I want to turn logic against quasi-logic and offer a logical analysis of this quasi-logical argument. Later on, I hope to show that such a logical analysis provides part of a profile of dialogue; that it constitutes part of the dialectic, but not all of it.

What would a logical analysis of the long lecture argument amount to? Generally, a logical analysis of an argument consists of two parts: a reconstruction and a critical evaluation. To reconstruct the argument, we notice that the argument gives the impression of being tightly reasoned and logical. We therefore try to reconstruct it as a logical derivation.[iii] One thing we need to attend to is the occurrence of a suppositional subargument that needs to be put into a more explicit format. As a first line of proof one may enter:

(1) I hate to hear long lectures.

(This may, in context, be taken to constitute a fact.)
As an unexpressed (and unproblematic) premise we may add:

(2) For all X, if I hate X, X is a bad thing for me.

And from this we may conclude:

(3) Hearing a long lecture is a bad thing for me.
As another unexpressed, empirical but, I think, unproblematic premise we may state:

(4) Whatever type of lecture I like, I get to hear it.
Then the argument introduces a supposition:

(5) Suppose: I like long lectures.
Given this supposition we may conclude, using (4):

(6) I get to hear a long lecture.
Whence, by (3) (maintaining the supposition):

(7) I get what is a bad thing for me.
By conditionalization we may retract the supposition to obtain:

(8) If I liked long lectures, I would get what is a bad thing for me.
For the sake of argument, let us pretend that the following premises are also acceptable:

(9) If I liked long lectures, that would do me no good.

(10) On account of my dislike of long lectures, no evil will befall me.

Thus, on balance, if I liked long lectures things would get worse as compared to my present state of dislike of them. Hence, let us presume that the calculus of rational sentiments now warrants the conclusion:

(11) I may be glad not to like long lectures.[iv]

One may expand the reconstruction so as to reach the conclusion that “I may rejoice in hating long lectures”, but the weaker conclusion will do. Passing to the evaluation stage, we first notice that the argument is less of an innocent conundrum than it might have been supposed.

What is worrying about the argument is its generalizability to a type: self-gratulatory argument. Given the present analysis, it seems that for all matters where choice is ethically neutral, a matter of taste, we have, no matter our predilections, special reason to be glad to have exactly those preferences we happen to have. If you hate red funiture, you should be glad that you hate red furniture.

Otherwise, you might have been buying loads of red furniture; and you so much hate the stuff! Yet the same holds for those that love red furniture: they have reason to rejoice in their preferences as well. This looks terribly suspect; you’re OK and I’m OK. Can it be? One logician’s solution, given this analysis, would be to point at line (8):

(8) If I liked long lectures, I would get what is a bad thing for me.

This line is obtained from the preceding suppositional argument by the rule of conditionalization. But conditionalization does not warrant the modalities (“liked” and “would”) that were introduced at this line. A proper application of conditionalization yields instead:

(8*) If I like long lectures, then I get what is a bad thing for me.

This proposition, which, by the way, follows immediately from premise (1), is much weaker. It is too weak to carry the rest of the argument. To reach the verdict that I may rejoice in my dislike of long lectures one needs to have recourse to the original modal version that tells us that I would be worse off in counterfactual situations in which I liked long lectures.

Alternatively, if one wants to save the full modal version of (8), one needs a modalized version of suppositional argument to support it. In a modalized suppositional argument reference to the preceding nonmodal part may be blocked. There are two such references: use is made of (3) and of (4). The use of (4) might be saved by giving this premise a modal formulation:

(4*) Whatever type of lecture I might like, I would get to hear it.

This modalized version of the premise will certainly not be canceled by the introduction of the counterfactual supposition that I  like long lectures. The use of (3), however, cannot be saved in this way. In order to establish the following proposition:

(3*) Whatever type of lecture I might like, hearing a long lecture would be a bad thing for me.

One would have to push up the modalization to its premises (1) and (2). For (2) this may be unproblematic but the modalization of (1) yields a problematic premise:

(1*) Whatever type of lecture I might like, I would hate to hear long lectures.

This is problematic, for its seems plausible that if I liked long lectures I would not hate to hear them.

Thus, following the gist of this logical analysis, it seems that the argument goes wrong somewhere, but we cannot tell for sure where it goes wrong. Now this result is not really spectacular: one could have suspected beforehand that something was rotten. What the logical analysis adds is a more precise insight in the ways the argument goes wrong; or better, in the ways it might go wrong. Proposition (8) must be either modalized or unmodalized. If it is unmodalized the trouble arises in the last part of the argument. If it is modalized then either the suppositional part must be modalized as well, or the application of the rule of conditional proof would be in error. Then, if the suppositional proof is modalized, one needs to modalize (3) to avoid a fallacious reference in the now modalized suppositional proof. Finally, if (3) is modalized one needs to modalize premise (1) to restore validity in the first part of the argument. But this leaves us with a problematic premise.

From the logical analysis one may extract a pattern of dialectic moves and countermoves, a kind of profile of dialogue. Together these moves determine a strategy for the critic of the self-gratulatory argument. First, corresponding to the reconstruction part of the logical analysis, the critic must try to get to an agreement on a more precise understanding of the argument. In this phase the critic may ask the proponent to reformulate parts of the argument in a clarifying way, but she may also, more actively, propose reconstructions of her own. Of course, this dialectic process may lead to an understanding of the self-gratulatory argument different than the present reconstruction. In that case the rest of the dialectical process will also be quite different from what follows. But let us assume that an agreement on the present reconstruction can be obtained. Corresponding to the evaluation part of the logical analysis, the critic should then, in the second phase, go on to ask whether proposition (8) is to be understood as modalized or as unmodalized. If the answer is that it is to be understood as unmodalized, then the critic is to turn to criticism of the last part of the argument. If the answer is that it is to be understood as modalized, then she should point out that the suppositional part of the argument must be understood as modalized as well. Once this much has been granted she may go on to push the modalization upward over (3) to (1). Finally the critic may point out that the modalized version of (1) is highly problematic.[v]

We would not speak of a quasi-logical argument if we did not think there was something wrong with it, and this presentiment, in the present case, was borne out by the strategy displayed. Without going as far as to claim that we here have a winning strategy for the critic of the self-gratulatory argument (for one thing, we did not check on other possible reconstructions), there can be no doubt that the proponent of that argument gets driven into a corner from which it is hard to escape. Such a strategy I shall call a strong critical strategy (namely, a strong strategy for the critic).
Using a strong critical strategy against a quasi-logical argument will most likely put the proponent at fault. But the strategy need not tell which particular step in the argument is the one to blame. Generally, in a particular discussion, a tournament, in which the critic uses the strategy, the fault will be pinpointed at a particular spot. But then it can hardly be said that the fault was there to start with; rather it seems that the fault was constructed to be right there by the outcome of the dialectical process. The situation is quite analogous to that so aptly described by Richard Whately as he points out the indeterminate character of some fallacies. These are situations where a fallacy has been committed, but you cannot tell which fallacy it is.[vi]
To sum up the lessons drawn from this first example: a language user confronted with a quasi-logical argument is not without means of defense. In order to convince her adversary that his argument fails, she may solicit reconstructions of the argument and offer reconstructions herself, until a picture is obtained that is sufficiently clear to pursue a strategy of detailed criticism and evaluation. In this type of defense the critic generally goes beyond the stance of pure critical doubt to engage herself actively in discussing merits and demerits of parts of the argument. Thus part of the profile of dialogue associated with quasi-logical arguments consists of these two types of moves: the reconstructive and the evaluative.

2. The New Rhetoric’s Idea of the Quasi-Logical
What may be surprising, or even disquieting, is that in stressing the importance of applied logic the present approach might seem to run counter to that of The New Rhetoric. Since The New Rhetoric constitutes the locus classicus for the concept of quasi-logical arguments, and since I have no intention of belittling my indebtedness to Perelmans and Olbrechts-Tyteca, it will be proper to shortly investigate the New Rhetoric’s notion of the quasi-logical and to see where the differences lie between their approach and mine. According to The New Rethoric, quasi-logical arguments avail themselves of techniques of formal demonstration in a context of informal argumentation. I quote the beginning of its chapter on quasi-logical arguments (translation in the notes):

Les arguments que nous allons examiner dans ce chapitre prétendent à une certaine force de conviction, dans la mesure où ils se présentent comme comparables à des raisonnements formels, logiques ou mathématiques. Pourtant, celui qui les soumet à l’analyse perçoit aussitôt les différences entre ces argumentations et les démonstrations formelles, car seul un effort de réduction ou de précision, de nature non-formelle, permet de donner à ces arguments une apparence démonstrative; c’est la raison pour laquelle nous les qualifions de quasi logiques. (1970: Section 45, p. 259)[vii]

One more quote brings out a characteristic feature of many quasi-logical arguments:
… l’accusation de commettre une faute de logique est, elle-même, souvent, une argumentation quasi logique. On se prévaut, par cette accusation, du prestige du raisonnement rigoureux. (1970: Section 45, p. 260)[viii]

The authors discuss several ways in which the exploitation of formal demonstration in a context of informal argumentation can be accomplished. For instance, the arguer may present as a formal contradiction what is merely an informal, perhaps a pragmatic, incompatibility.[ix]

As a second example, I mention the informal division of a domain, which may be exploited, quasi-logically, as a basis for a completely rigorous constructive or destructive dilemma.[x]

Perelman and Olbrechts-Tyteca also discuss several means to fend off these quasi-logical arguments. Thus, where the claim is to have shown up a formal contradiction, one may try to show that it is merely a matter of incompatibility, that is, that one’s opponent has reduced or simplified the meaning of some statements in order to assimilate the system under attack to a formal system.[xi]  In the case of a quasi-logical dilemma, they point out the possibility of converting it into a counterdilemma; this would amount to answering a quasi-logical argument by a quasi-logical argument. But they also mention a method of more general application that allows the critic to deconstruct a dilemma argumentatively: this method has the critic allege qualifications of time and other nuances that permit him, argumentatively, to slip between the horns of the dilemma (1969: 238; 1970, Section 56, p. 321).[xii]

The authors do not mention the possibility of detailed logical criticism, as embedded in a strong critical strategy, of a quasi-logical argument. It is not unlikely that in their view such criticism would be itself quasi-logical. On the other hand, at certain junctures they seem not to object to answering a quasi-logical argument by another one. So perhaps they would not object to the type of responses given in a strong critical strategy. Conversely, our profile of dialogue may be enriched by the inclusion of a branch that offers the Perelmanian option of answering a quasi-logical argument by a quasi-logical counterargument. The counterdilemma, for instance, may be looked upon as an invitation to retract the original dilemma without having to go through a detailed logical analysis of either dilemma. As such it is not unreasonable.[xiii]
Another type of move that is rightly stressed by Perelman and Olbrechts-Tyteca is that of introducing terminological clarifications and qualifications. These moves form an important part of the dialectic; we must assume them to be prominent in the discussion phase in our profile that corresponds to the reconstruction part of the logical analysis.
Yet another type of move is suggested by the aforementioned characterization of certain quasi-logical arguments as taking advantage of the prestige of rigorous thought. In as far as an arguer tries to intimidate his opponent by the use of vocabulary or other resources taken from logic and mathematics (or, for that matter, from any other prestigious field), where these resources have no real role to play, he may be charged with ad verecundiam. So we may add a move to the profile that introduces this type of charge.

Now that we have noted these valuable contributions, we should not refrain from mentioning two rather worrisome features of the account in The New Rhetoric. The first of these concerns the question whether quasi-logical arguments are ultimately to be evaluated negatively. And if so, whether ordinary logical deductive arguments, too, would have to be called quasi-logical in most cases. On these points The New Rethoric leaves us somewhat in the dark.[xiv]
On the view I want to defend, we can hold on to a distinction between the logical and the quasi-logical; the first being a positive and the second being a negative term of evaluation. But who is to decide whether an argument is logical or not?[xv]
The point of view I want to take on this question is an immanent dialectical one. That is, ultimately, the status of an argument must be decided in discussion, by the participants themselves. Dependent on that outcome the argument is reconstructed as valid, as doubtful, as erroneous, as a blunder, or even as a fallacy. An argument that is presented as deductive and logical may be reconstructed as logical or as quasi-logical, and in the latter case as doubtful, erroneous, or fallacious. But such predicates may also be applied to an untested, unreconstructed argument – an argument “on the hoof”, as John Woods would say (1995: 187). In this case they must be taken as a preliminary verdict by which the speaker indicates a presumption that, after having been reconstructed and discussed, the argument would most likely be thus designated. A preliminary verdict can be given by an outsider from a spectator’s perspective, or it can be given by a participant, before the show begins. In particular, the predicate “quasi-logical” indicates that the speaker, whether a spectator or a participant, presumes the profile of dialogue to contain a strong critical strategy.

From this point of view, there is no need to designate all arguments that make use of resources taken from logic or mathematics as “quasi-logical”. Some of these are far better designated as “logical”. The latter predicate, as used in a preliminary verdict, would indicate that the speaker expects the argument to prove its mettle when tested by a critical opponent. So these two predicates express different expectations as to what the profile of dialogue contains. It is true that on a abstract level the profiles of dialogue for logical and for quasi-logical arguments will contain the same types of move; but only the latter profiles will contain a strong critical strategy once specified.[xvi]
The second worrisome feature of the account in The New Rhetoric is that the authors base their treatment on a dichotomy between a realm of formal demonstration and a realm of informal argumentation.[xvii] As I see it, this whole dichotomy has been misconceived, whether we interpret “formal” as “formalized” or merely as “rigorous”. For reasons of time, I shall skip that part of my paper, which, as you may have noticed, stands in danger of presenting a quasi-logical dilemma.[xviii]

3. Are Quasi-Logical Arguments Fallacies?
Now that we have reviewed the treatment in The New Rhetoric, we may return to the construction of a profile of dialogue. But first I want to take up the issue of whether quasi-logical arguments are fallacies. The answer of course depends on one’s theory of fallacy, but if I were to survey them all this would become a long lecture, indeed. Let me therefore announce that in the present context I take fallacies to constitute transgressions of the rules of what may either be called “persuasion dialogue” or “critical discussion”. Acts that conform to the rules of dialogue, but are strategically inferior are not fallacies, but errors or blunders. The point of the distinction is that one may see it as a goal of critical discussion, subsidiary to its primary goal of conflict resolution, that the arguments that are put forward in it are critically tested. This means that in good dialogue both outcomes must be possible: sometimes an argument will pass all tests and sometimes an argument will fail. Consequently, putting forward a bad argument is not by itself a fallacy. It need not be unreasonable. Just as it is not by itself unreasonable to lose a discussion or a part of it, but only to fail to admit the dialectical consequences of one’s loss, so it is not by itself unreasonable to argue quasi-logically, but only to fail to admit that one has done so, once the flaw has been exposed.[xix]

4. A Profile of Dialogue
A profile of dialogue for quasi-logical and other arguments can now roughly be sketched as folows (see Figure 1). It is a profile that pertains to argument criticism in general; but, by means of an example, I hope to show presently how it may be applied to criticism of arguments that claim to be logically tight.[xx]
It is supposed that the argument is presented within a context of persuasion dialogue. Most simply, let there be two parties Wilma and Bruce. Suppose that Wilma has advanced a thesis and that Bruce has challenged this thesis and that Wilma has put forward an argument in order to defend her thesis. The argument can be either simple or complex. A simple argument contains only one premises/conclusion structure, a complex argument may contain a whole tree of such structures as well as suppositional parts.

I now want to see what dialectical moves should be available for Bruce to react to the argument. One reaction for Bruce would be to accept the argument as adequate and to retract the challenge. Other reactions are more or less critical.
The least critical of these other reactions would be to renew the challenge. Bruce may simply declare not to have been convinced by the argument. But in order that the dialectic process move forward, it is then incumbent upon Bruce to indicate precisely those steps in the argument that failed to convince him. This gives Wilma the opportunity to expand the argument precisely at those turns where expansions are required to achieve her particular goal in the dialogue, namely to convince Bruce. Notice that in the case of simple arguments this part of the profile reduces to moves of tenability criticism (are the reasons given themselves acceptable?) and of connection criticism (is the reason adequate to support the thesis?).
A number of more critical reactions for Bruce are grouped under the heading of active criticism. In these branches of the profile, Bruce takes upon himself a burden of proof to show that the argument, though perhaps not unreasonably proposed at this point of dialogue, is ultimately wrong, mistaken or insufficiently weak in some way or other. In this branch one finds counterarguments (including quasi-logical ones) and argument criticisms of various sorts.
Finally a third type of reaction for Bruce would be to put up a fallacy criticism. Bruce now denies that Wilma’s argument might be reasonable. On the contrary, it is claimed that the argument is inadmissible. That is, that it infringes such rules for persuasion dialogue (including rules of logic) as obtain in the company to which both disputants belong. The retraction Bruce is after is not the regular retraction that takes place on the ground level dialogue, but a retraction of the argument as an argument that never should have been put forward in the first place.[xxi]

5. An Example: The Immortality Argument
To fill out this rather sketchy profile a little further, let us contemplate another example. In it Wilma and Bruce discuss a proof of immortality that has often proved to be hard to disentangle. After each move I shall indicate its place in the profile.

Wilma: We, human beings, are immortal. [thesis]
Bruce: How come? [challenge]

Wilma: This can be proved by sharp logical reasoning. For suppose we were mortal. In that case a good question to ask would be: shall we remain mortal? There are exactly two cases to consider: either we shall remain mortal for ever or we won’t. Suppose we shall remain mortal for ever. In that case we shall remain for ever. So in that case we must be immortal. On the other hand, supposing that we shall not remain mortal for ever, we must become immortal at some time in the future. But whosoever will become immortal at some time in the future, happens to be immortal right now. Consequently, both cases lead to the conclusion that we are immortal right now. The supposition that we would be mortal, therefore, has as a consequence that we happen to be immortal. From which we may conclude, by impeccable logic, that we must be immortal. [argument presented as “proof”: see Figure 2 for a survey]
Bruce: First of all I want to object to calling this argument a proof and to your calling your own logic impeccable. You have not indicated what special features would justify one to speak of a proof. For instance, you have not mentioned any axiomatic theory such that your argument would be a proof within the context of that theory. So, please, withdraw the claim to have provided a proof and admit that your logic still needs to be tested in critical discussion. [fallacy criticism: ad verecundiam]
Wilma: O.K. Let us call it an argument. [retraction of “proof”] But is there anything wrong with it? [upholding the argument]
Bruce: Let me see. Your argument sounds rather fishy. Many people will object to the last part, where, from the result that the supposition that we would be mortal leads to the consequence that we are immortal, you conclude that we are immortal. Your conclusion really seems conjured up out of a hat. But, fortunately for both of us, I studied enough logic to see there is nothing wrong with this last step. It is a reductio ad absurdum. So there must be some other mistake in your argument. Could I put my finger on a false dilemma? I suppose you realize that, for your case-splitting to be exhaustive, it must be presumed that we are either all equally mortal or all equally immortal? [active criticism: logical analysis, reconstructive phase]
Wilma: Yes. But alternatively one could replace the “we” in the argument by each of our proper names in turn: Frans, Rob, Tony, Charley, and so on. Thus the argument would show each of us, separately, to be immortal, and the case-splittings would all be safe. [alternative reconstruction]
Bruce: Uh, well, let us look at these cases. Aha! There you are! You say that if we shall remain mortal, we shall remain. The first “remain” is a linking verb, the second is an intransitive verb. That cannot be right! [active criticism: logical analysis, evaluative phase]
Wilma: Please pay attention to the thought rather than to the words. What I meant to say is that if we shall remain mortal for ever, that is, if we shall be mortal at any future point of time, then at any future point of time we must be there to be mortal. We can’t be mortal when we are dead. [back to reconstructive phase]
Bruce: I see. Now I’m confused. I thought to have spotted the flaw. But this was a kind of red herring in your argument, was it not? It looked like a flaw , but it wasn’t. I’d say your presentation is somewhat at odds with our rules of dialogue – the tenth rule of pragma-dialectics to be exact.[xxii] How can one pay attention to the thought if the words are jumbled? [fallacy criticism] Anyhow, I now grant the first horn of your dilemma. But in the second horn I think I can spot a problematic premise. You say that anyone who will become immortal at some time in the future, happens to be immortal right now. But take the case of Hercules, the ancient hero. At the end of his life he was adopted by the immortal gods; so during his life, as he was still a mortal being, it was true to say of him that he would become immortal at a certain time in what was then the future. But, before the gods granted him this great favor, he was not immortal yet. Wouldn’t that disprove your premise? [active criticism: counterargument]
Wilma: I’m sorry if my way of expressing has been misleading. Now as to the apotheosis of Hercules, you are right that there is a distinction to be made. On the one hand, “immortality” may be understood as an intrinsic property, such that for whoever has the property dying is impossible. This was the property the gods conferred on Hercules. But I meant “immortality” to be understood as an extrinsic property, that is, simply as the property one has if one will, in fact, live forever (even though death may in all eternity remain an unrealized possibility). Now if at some time in the future one of us has the property of living at that time and forever after, he or she must necessarily live through all moments of time from the present moment onwards. That is, that person must right now be immortal in the extrinsic sense. [back to reconstructive phase]
Bruce: I never was aware of this ambiguity in the meaning of “immortal” But the distinction seems almost evident now that we have delved so deeply into this argument. Yes, I suppose I can see it your way. [idem]
Wilma: Since now you have checked all parts of the argument, isn’t it about time to get to the concluding stage of this discussion as well as of this rather lengthy lecture in which our discussion is embedded? Are you prepared to withdraw your critical doubt with respect to my thesis? [walks to the door as she is getting to the concluding stage]
Bruce: Now wait a minute! If your thesis is right and we are immortal, there is plenty of time, so why hurry? [ad hominem: you don’t practise what you preach]
Wilma: Because you cannot beat my argument, dumbo. [abusive ad hominem]

We must leave Bruce and Wilma right in the middle of this altercation which, by the way, provides us with a clear-cut example of a dialectical shift.[xxiii] After all, I hate long lectures.

Figure 1 A Profile of Dialogue

6. Summary and conclusions
Summing up: In this lecture I wanted to discuss the dialectical moves that are appropriate in a critical discussion of a quasi-logical argument. Two examples of quasi-logical arguments have been presented for the purpose: the self-gratulatory argument and the immortality argument. Though the dialectical analysis of both of them had to be rather sketchy, I hope to have raised some interest in the dialectical study of arguments by means of the specification of profiles of dialogue. Ultimately, as I see it, the study of profiles is to help us construct rigorously formulated systems of formal dialectics (as in Walton and Krabbe 1995: Ch. 4); but I have not touched upon these. In the present case of quasi-logical arguments the dialectic was seen to link up closely with logical analysis, from which strong critical strategies could be derived, but we have also profited from the rhetorical point of view expounded in The New Rhetoric.

Adhering to the pragma-dialectical concept of fallacy, I did not want to say that all quasi-logical arguments are fallacious. Moreover, I did not envisage a theory of dialogue that would in all cases be able to decide on such matters as quasi-logicality or fallaciousness beforehand. In many cases, the theorist will have to refrain from anything more than a preliminary judgment. According to the immanent dialectical approach it must often be left to the disputants themselves to decide these matters. But their decision is not arbitrary. In their discussions the disputants are supposed to be guided by rules of dialectics that are accepted by the company of discussants to which they belong. The empirical and normative study of these rules is the task of the dialectician.

NOTES
[i] The present paper contains the literal text of my ISSA-conference keynote, read on June 18th, 1998. I hope the reader will be willing to excuse a number of peculiarities of style which are due to its having been written for the ear rarther than for the eye. Some notes, references, captions, and figures have been added. A summary of the paper is contained in the last two paragraphs. I want to thank the board of the International Society for the Study of Argumentation (ISSA) for their invitation; my Groningen colleagues of the Promotion Club Cognitive Patterns (PCCP), who were the victims of a first try-out presentation, for a number of helpful suggestions; and David Atkinson for some prompt and apposite linguistic advice. Finally, I dedicate this piece to the memory of my brother, Theo, with whom I first invented and discussed the argument for immortality.
[ii] This rather sketchy definition is, I think, all that is needed to delineate my topic. The term ‘unwarrantedly’ introduces a normative element to be fleshed out from various perspectives.
[iii] Whether one is a deductivist or not, this way to reconstruct an argument seems perfectly legitimate when the mode of presentation of the argument invites us to see it as a rigorous deduction.
[iv] I doubt whether anything like a satisfactory ‘calculus of rational sentiments’ exists, but, anyhow, some sentiments are deemed more reasonable than others in certain contexts. We argue about these things with an – often implicit – appeal to “feeling rules” (Hochschild, 1979). As Arlie Hochschild wrote: In common parlance, we often talk about our feelings or those of others as if rights and duties applied directly to them. For example, we often speak of “having the right” to feel angry at someone. Or we say we “should feel more grateful” to a benefactor. We chide ourselves that a friend’s misfortune, a relative’s death, “should have hit us harder,” or that another’s good luck, or our own, should have inspired more joy. We know feeling rules, too, from how others react to what they infer from our emotive display. Another may say to us, “You shouldn’t feel so guilty; it wasn’t your fault, You don’t have a right to feel jealous, given our agreement.” (p. 564) One may agree with this idea – that we can discuss the rationality, or the appropriateness, of emotions, sentiments, or feelings – without committing oneself to the view that emotions are judgements (Solomon, 1980). The rationality of emotions as patterns of salience has been
discussed by Ronald de Sousa (1980).
[v] If the proponent of the original argument refuses to move along with these latter criticisms, say if he holds on to a step that derives a modalized from an unmodalized proposition, the critic may need to go into detail to demonstrate the invalidity of that particular step. Here she may have recourse to several methods in order to convince the proponent: counterexample, logical analogy, and formal analysis. In some cases, if the proponent can be held to certain rules of logic on account of which the invalidity should have been obvious, he may be brought to admit to have committed a fallacy (cf. my 1995). We shall get back to fallacies later.
[vi] Whately writes: if a man expatiates on the distress of the country, and thence argues that the government is tyrannical, we must suppose him to assume either that ‘every distressed country is under a tyranny,’ which is a manifest falsehood, or, merely that ‘every country under a tyranny is distressed,’ which, however true, proves nothing, the Middle Term being undistributed. (1836: 149-50) According to Whately, a fallacy has been committed, but you cannot tell which fallacy it is.
[vii] Translation: The arguments we are about to examine in this chapter lay claim to a certain power of conviction, in the degree that they claim to be similar to the formal reasoning of logic and mathematics. Submitting these arguments to analysis, however, immediately reveals the difference between them and formal demonstrations, for only an effort of reduction or specification of a nonformal character makes it possible for these arguments to appear demonstrative. This is why we call them quasi-logical. (Perelman and Olbrecht-Tyteca 1969: 193)
[viii] Translation: … the charge of having committed a logical error is often itself a quasi-logical argument. By making this charge, one takes advantage of the prestige of rigorous thought. (1969: 194)
[ix] Thus the arguer may pretend to have given a proof by the celebrated logical pattern of reductio ad absurdum, whereas he has done no more than raise some objections that might be answered.
[x] Thirdly, informal definitions that cover terminological manipulations may be presented as formal stipulations without theoretical content which, moreover, are taken to warrant unscrupulous substitutions by Leibniz’s Law.
[xi] Perelman and Olbrechts-Tyteca write: Aussi, un des moyens de défense qui sera opposé à l’argumentation quasi logique faisant état de contradictions sera de montrer qu’il s’agit non de contradiction mais d’incompabilité, c’est-à-dire que l’on mettra en évidence la réduction qui seule a permis l’assimilation à un système formel du système attaqué, lequel est loin de présenter, en fait, la mème rigidité. (1970, Section 46, p. 263) Translation: Therefore one of the means of defense to be used against the quasi-logical argument which claims a contradiction is to show that it is not a matter of contradiction but of incompatibility. In other words, one will display the reduction which alone has made possible the likening to a formal system of the system under attack, which in fact does not exhibit the same rigor. (1969: 196).
[xii] As to the example of Note 10: Perelman and Olbrechts-Tyteca do not explicitly discuss any techniques to ward off uncongenial definitions (except for the somewhat dubious charge of tautology), but it is clear that they consider definitions as argumentative, and therefore in principle as a proper focus for critical objections in a context of informal argumentation.
[xiii] This is not to say that, in critical discussion, it could be reasonable to negotiate and to ‘trade off dilemma’s’. Rather the counterdilemma is an expedient to convince the other party that the original dilemma does not hold water.
[xiv] On the one hand quasi-logical arguments are not officially designated by any negative terms such as ‘error’, ‘flaw’, or ‘fallacy’, and neither would one expect such a verdict from The New Rhetoric. On the other hand the term ‘quasi-logical’ by itself has a negative ring, and in discussing such arguments the authors strike a particularly critical note. An argument’s claim to be similar to the formal reasoning of logic and mathematics would hardly ever be justified. This leaves us wondering. A situation that is aggravated by the rather puzzling fact that many examples in The New Rhetoric are based on valid logical schemata such as the constructive dilemma, be it that their application remains somewhat doubtful. This makes one wonder whether there is any distinction between ordinary logical deductive arguments and quasi-logical arguments.
[xv] For an argument to be designated as quasi-logical, it is not sufficient that its mode of reasoning be taken from logic or mathematics. As stated in the introduction, it must also be the case that the transfer is ‘unwarrented’. But who decides whether this is the case or not?
[xvi] On an abstract level, a profile of dialogue, as in Figure 1, merely shows what possible types of moves are available for the disputants. Once the general scheme has been applied to a specific thesis, one obtains a survey of possible specific moves, from which strategies for either party may be selected.
[xvii] This dichotomy might explain the authors’ resistance to the idea of admitting a group of arguments that are plainly logical and not quasi-logical. So-called plainly logical arguments, they might want to say, would illegitimately treat a context of informal argumentation as if it were a context of formal demonstration.
[xviii] Braving such risks, I here present the argument that had to be skipped: The core of the trouble lies in the concept of “formal demonstration”. On the one hand “formal demonstration” may be taken to refer to formalized axiomatic deductions, that is, deductions in a formal language using a fixed set of rules of inference. But the construction of formal languages and formal deductions is much too specialized an activity for it to have such an impact on informal argumentation as Perelman and Olbrechts-Tyteca ascribe to formal demonstrations. For one thing, it is not the formalization of logic and parts of
mathematics that is responsible for the prestige of rigorous thought. Nor do the ideas of contradiction, definition, identity, dilemma, etc. originate from these formal systems. Also, if formalized axiomatic deductions were the standard that quasi-logical arguments exploit, it would remain a mystery how there could have been quasi-logical arguments before Frege.
On the other hand, one may interpret “formal demonstration” as “rigorous demonstration” or “proof”. Then we get to a concept of formal demonstration that very well explains the ad verecundiam character of many quasi-logical arguments. This is something to be aware of, and Perelman and Olbrechts-Tyteca were right to stress it. But there is simply no dichotomy. Logical and mathematical proofs are just one of a kind with informal arguments. Proofs cannot be taken as absolutes: what counts as a proof for one person may not count as a proof for someone else. To call an argument a proof announces a surplus value above the more mundane types of informal argument, such as being part of an (informal) axiomatic set-up, but the character of this surplus value may vary in different contexts. (This is explained more fully in my 1997.) In some cases the label “proof” may just constitute an ad verecundiam or ad baculum ingredient of one’s argument. Thus, what is presented as a rigorous proof is a potential object of analysis, of reconstruction and evaluation. In this proofs do not differ from other types of informal argumentation. Misuse of what appears to be schematically correct schemata is not excluded in the area of demonstration, as is witnessed by the logical paradoxes and the existence of flawed proofs. We may conclude that if “formal demonstrations” are interpreted as rigorous (but informal) proofs, quasi-logical arguments could occur just as well within the context of formal demonstrations as in the context of juridical, philosophical or everyday argumentation. Hence it would be impossible to explain quasi-logical arguments as attempts to emulate formal demonstrations in a context of informal argumentation. Thus either interpretation of the term “formal demonstration” lands us in difficulties. This poses a (hopefully not quasi-logical) destructive dilemma for the whole idea of basing the concept of quasi-logical argument on a dichotomy between informal argumentation and formal demonstration.
[xix] The cases where quasi-logical arguments are fallacies are those that may be shown to fulfill some extra conditions, among which figures pre-eminently that they must transgress the operative rules of dialogue. Which quasi-logical arguments are fallacies depends, then, on the rules of dialogue that hold in the context.
[xx] Profiles of dialogue are tree-shaped descriptions of options and possible sequences of moves in reasonable dialogue. Here “reasonable” does not imply that no fallacies are committed, but that fallacies and challenges of fallacies are adequately handled within the dialogue. The method of profiles aims at getting a survey of all the different ways a reasonable dialogue of some type could proceed. It was applied by Douglas Walton in his discussions of the fallacy of many questions (1989a: 68, 69; 1989b: 37, 38) and in several of his later books on fallacy theory (1995: 22-26; 1996: 150-54; 1997: 253-55). Cf. also my 1992 and 1995.
[xxi] The subject of retraction is a tricky one. Surely, one should admit reasonable retractions: retractions of fallacious moves of course, but also ground level retractions, say of standpoints a disputant has been unable to defend in a satisfactory way. On the other hand, persistent retraction of each of one’s commitments in dialogue would make reasonable discussion all but impossible. The problem of where and how to draw the line is one of the main themes in Walton and Krabbe 1995.
[xxii] ‘… Rule 10 for a critical discussion runs as follows: A party must not use formulations that are insufficiently clear or confusingly ambiguous and he must interpret the other party’s formulations as carefully and accurately as possible.’ (Van Eemeren and Grootendorst, 1992: 196)
[xxiii] In fact, Bruce’s tu quoque and Wilma’s abusive ad hominem secured them a fast cascading down into a quarrel. See Walton and Krabbe 1995, Sections 3.3 and 3.4, esp. pp. 105-7 and 111-12. A closer look at Proposition 6 (Figure 2) would have been more profitable

REFERENCES
De Sousa, Ronald (1980). The Rationality of Emotions. In: Amélie Oksenberg Rorty (ed.), Explaining Emotions, Berkeley, Los Angeles, and London: University of California Press, pp. 127-51, Ch. 5.
Hochschild, Arlie Russell (1979). Emotion Work, Feeling Rules, and Social Structure. American Journal of Sociology 85 (3), 551-75.
Krabbe, Erik C.W. (1992). So What? Profiles for Relevance Criticism in Persuasion Dialogues. Argumentation 6, 271-83.
Krabbe, Erik C.W. (1995). Can We Ever Pin One Down to a Formal Fallacy? In: Frans H. van Eemeren, Rob Grootendorst, J.
Anthony Blair and Charles A. Willard (eds.), Proceedings of the Third ISSA Conference on Argumentation (University of Amsterdam, June 21-24, 1994) II: Analysis and Evaluation, Amsterdam: Sic Sat, International Centre for the Study of Argumentation, pp. 333-344. Also in: Theo A. F. Kuipers and Anne Ruth Mackor (eds.), Cognitive Patterns in Science and
Common Sense: Groningen Studies in Philosophy of Science, Logic, and Epistemology, Amsterdam and Atlanta GA: Rodopi, 1995, pp. 151-64, and in: Johan van Benthem, Frans H. van Eemeren, Rob Grootendorst and Frank Veltman (eds.), Logic
and Argumentation, Amsterdam, etc.: North-Holland, 1996, pp. 129-141.
Krabbe, Erik C.W. (1997). Arguments, Proofs, and Dialogues. In: Michael Astroh, Dietfried Gerhardus, and Gerhard Heinzmann (eds.), Dialogisches Handeln: Eine Festschrift für Kuno Lorenz, Heidelberg: Spektrum, Akademischer Verlag, pp. 63-75.
Perelman, Chaïm and Lucie Olbrechts-Tyteca (1969). The New Rhetoric: A Treatise on Argumentation (Translation by John Wilkinson and Purcell Weaver of Perelman and Olbrechts-Tyteca, Traité de l’argumentation: La nouvelle rhétorique, Paris: Presses Universitaires de France (2 vols.), 1958). Notre Dame, IN, and London: University of Notre Dame Press.
Perelman, Chaïm and Lucie Olbrechts-Tyteca (1970). Traité de l’argumentation: La nouvelle rhétorique, 3rd ed. Brussels: Editions de l’Université de Bruxelles. First edition: Paris: Presses Universitaires de France (2 vols.), 1958.
Solomon, Robert C. (1980). Emotion and Choice. In: Amélie Oksenberg Rorty (ed.), Explaining Emotions, Berkeley, Los Angeles, and London: University of California Press, pp. 251-81, Ch. 10.
Van Eemeren, Frans H. and Rob Grootendorst (1992). Argumentation, Communication, and Fallacies. Hillsdale, NJ, Hove, and London: Lawrence Erlbaum Associates.
Walton, Douglas N. (1989a). Question-Reply Argumentation. New York, Westport, CT, and London: Greenwood Press.
Walton, Douglas N. (1989b). Informal Logic: A Handbook for Critical Argumentation. Cambridge, etc.: Cambridge University Press.
Walton, Douglas N. (1995). A Pragmatic Theory of Fallacy. Tuscaloosa, AL, and London: The University of Alabama Press.
Walton, Douglas N. (1996). Arguments from Ignorance. University Park, PA: The Pennsylvania State University Press.
Walton, Douglas N. (1997). Appeal to Expert Opinion. University Park, PA: The Pennsylvania State University Press.
Walton, Douglas N. and Erik C.W. Krabbe (1995). Commitment in Dialogue: Basic Concepts of Interpersonal Reasoning. Albany, NY: State University of New York Press.
Whately, Richard (1836). Elements of Logic. New York: William Jackson. First edition 1826.
Woods, John (1995). Fearful Symmetry. In: Hans V. Hansen and Robert C. Pinto (eds.), Fallacies: Classical and Contemporary Readings, University Park, PA: The Pennsylvania State University Press, pp. 181-93, Ch. 13.

You should always try to avoid the use of clichés. (anonymous)

1. Clichés Don’t Grow on Trees – Introducing Clichés
Following the unprecedented growth and dissemination of information and the widespread access to it through the media, we are increasingly experiencing the use of clichés, old and new, unchanged and altered, famous and anonymous: ‘Life imitates art’, ‘All the world’s a stage’, ‘It’s a small world’, ‘Money talks’, ‘Time is money’, ‘Money does not grow on trees’, ‘Traduttore, traditore’, ‘Cherchez la femme’, ‘the man in the street’, ‘political correctness’, ‘I promise to love you until death do us part’, ‘Men and women are different: Vive la différence!’, ‘Elementary, my dear Watson’, and so on. These frequently recycled expressions are looked upon as unquestionable truths or at least as ‘le mot juste’ by many people.  Some, however, dismiss them as “clichés”.
This paper is devoted to clichés. Not to discard them, but to make some observations about their relevance to argumentation and their potential for miscommunication. Actually, we claim that certain clichés are crucial to argumentative discourse, and that their capacity for building arguments is closely linked to their liability to trigger divergent interpretations.
We propose a pragmatic and rhetorical approach to the concept of cliché and its functions in argumentation. This approach takes into consideration three major elements in the dynamics of clichés, the disregard of which may lead to misinterpretations:
1. there is no complete overlap between the form and the function of the lexical entitities that underlie clichés
2. many clichés exhibit a balance between a general scope and a specific focus on certain topoï for which a particular audience is expected to have a particular preference at a particular time in a particular context
3. there is an inherent tension between the explicit and the implicit functioning of a cliché in argumentation

Cliché is a word with a negative ring to it. When you say “This is a cliché” about an opinion voiced by a partner in conversation, you usually imply that s/he is yielding to popular unreflected opinion, that s/he is just repeating something constantly circulating in the mental marketplace of a certain discourse community. A cliché is then seen as a commonplace, the collective consensus speaking through the mouth of an individual without involving his/her own critical thinking (Lerner: 1956, Ricks: 1980). Maybe what a cliché stands for is not blatantly untrue to a rational observer, as a prejudice usually is. But it is still likely to be seen as a crude and simplified way of looking at things that deserve a deeper and less biased consideration.
It is not easy to come to grips with clichés because their form does not display any regular patterns, their structure is difficult to capture and their occurrences impossible to predict. Whether the coinage of clichés is ascribed to well-known or to anonymous sources, it is their distribution and frequency that eventually decides their subsequent evolution.
Generally speaking, cliché seems to be a rather elitist word. Popular wisdom is not likely to come as close to the truth as a well-educated, highly trained and critical mind, such as the typical academic intellectual. Cliché, with its derogatory value load, is a word that Plato, that outspoken critic of the masses, could have used. It would have come less natural to Aristotle with his respect for ‘doxa’, for tradition and general opinion. We will side with Aristotle on this issue and try to show that clichés fulfil no negligible role even in informed discourse, rational argumentation and creative problem-solving dialogue.

2. What’s in a Cliché? – Defining Clichés
Clichés are often defined as stereotypical forms that have been proliferating in many areas of life, such as art, philosophy, behaviour, and language. Whether we like them or not, they represent an important ingredient in verbal and non-verbal communication and are meant to establish or signal common ground. Our education is, after all, based on certain fixed patterns of thinking and speaking.
The origin of the word cliché can be traced back to the technical jargon of the French printing trade in the nineteenth century. It denoted a cast obtained by dropping a matrix face downwards upon a surface of molten metal on the point of cooling (Howard: 1986). It may be an echoic word since it imitates the plopping sound that the matrix made as it fell into its hot bath, which is rendered in English by ‘click’ and ‘clack’, as has been pointed out by Redfern, who adds:
Because of its origins, together with ‘stereotype’ in printing, and its later extension to photography, the term parallels the development of modern technology. Imitation, identical reproduction (cloning, before its time), such associations led on to the figurative meaning (because reproducibility entails wear and tear) of mechanized mental processes and textural fatigue. (1989: 8)
According to Redfern, “famous quotations become clichés when they are trivialized by inappropriate use, for example: ‘To be or not to be’, parroted when a footling decision has to be taken” (1989: 41). He considers that kitsch is “the twisting of clichés to non-productive ends” (1989: 61).

Some dictionary definitions tend to draw a fine line between cliché as a repetitive formula, and stereotype as a more negatively loaded and oversimplified evaluative formula and mental attitude. Thus the word cliché is defined as “a stereotyped expression, a commonplace phrase” by the Oxford English Dictionary (1989), “a hackneyed phrase or expression; also the idea expressed by it; a hackneyed theme or situation” by the Longman Webster English College Dictionary (1985), and “a form of expression that has been so often used that its original effectiveness has been lost” by the Cambridge International Dictionary of English (1995). The word stereotype is defined as “something continued or constantly repeated without change; a preconceived and oversimplified idea of the characteristics which typify a person, situation, etc.; an attitude based on such a preconception” by the Oxford English Dictionary (1989), as “somebody or something that conforms to a fixed or general pattern; esp. a standardized, usu. oversimplified, mental picture or attitude that is held in common by members of a group” by the Longman Webster English College Dictionary (1985), and as “disapproving (a person or thing that represents) a fixed set of ideas that is generally held about the characteristics of a particular type of person or thing, which are (wrongly) believed to be shared by all the people and things of that type” by the Cambridge International Dictionary of English (1995).
As can be seen in the definitions above, the most common connotations of the term stereotype are overwhelmingly negative, which is not always the case with clichés. Stereotypes are conceived of as subjective and prejudiced speech and thinking habits, as well as the label of disapproval given by a person to another person’s generalisations (Missimer: 1990). Unlike clichés, which apply mostly to verbal and visual expression, stereotypes are especially used in connection with human types, attitudes, as well as human perception and behaviour. In social psychology, the method of phenomenology has been used to highlight and account for various stereotyped classifications, which tend to perceive and evaluate people as specimens of a social type. However, Ichheiser (1949) provides a more complex and nuanced picture of social and psychological stereotypes. He emphasizes that, in spite of their predominantly negative evaluation, “the preformed stereotyped images about other people are certainly among the most important factors in the system of ‘collective representations’ necessary to guarantee a minimum of consensus for a group. They should not be lightly dismissed as ‘prejudices’” (1949: 34). We subscribe to his claim that classificatory stereotypes contain both elements of truth and elements of falsehood. After a reconsideration of the linguistic form of clichés in 3. below, we will argue in 4. below for a new way of redefining clichés in terms of the relation between their linguistic form, on the one hand, and their discursive and rhetorical structure, on the other.

3. Words Don’t Make the Cliché – The Linguistic Form of Clichés
In what follows we proceed to a reevaluation of the criteria usually applied to the definition and classification of clichés as distinct from idioms, euphemisms, and other more or less fixed linguistic expressions.
Clichés have been studied by scholars from different disciplines, such as linguistics, literary studies, psychology, sociology, and political studies, to name but a few. After examining the items in Partridge’s Dictionary of Clichés, Luelsdorff (1981) has analysed their phonological, syntactic and semanic features, and makes the distinction between clichés exhibiting nominal, verbal or sentence structures. Whereas Luelsdorff focuses on the strictly linguistic features of clichés, Howard (1986) takes a step further, by discussing various subtypes of clichés in terms of idiomaticity and distinguishes between non-idiomatic and idiomatic clichés.
Of more recent date is the research carried out by Gramley and Pátzold (1992) on so-called ‘prefabricated language’, namely multi-word units or lexical phrases, including clichés. They discuss the effects and functions of clichés in connection with an analysis of the wider category of fixed expressions. Like Redfern (1989), they claim that clichés fulfil an important social function and can be assigned even a positive role in those areas of human interaction where consciously thought-out language is unusual, if not inappropriate, such as funerals, disasters, the writing of references and testimonials.
Concerning the distinction between an idiom and a cliché, Gramley and Pátzold are quoting Brook’s conclusion: “Whether we call a phrase an idiom or a cliché generally depends on whether we like it or not” (1981: 14). This statement appears to confirm our own intuitions, even though it offers little enlightenment about the nature of clichés. It is precisely at this point that a discussion of the defining features of clichés should actually start. None of the studies mentioned above has succeeded in pinning down the properties that distinguish clichés from other related linguistic expressions such as idioms. In trying to establish linguistic parameters and comparative default features for clichés, the authors overlook the crucial fact that clichés can hardly represent a linguistic category, but rather a pragmatic and a rhetorical category. In other words, clichés cannot be treated and classified as syntactic and/or lexical entities, i.e. according to grammatical form and structure, but rather as functional elements made up of longer or shorter stretches of words, from one word to a whole utterance. This insight may help to explain why both idiomatic and non-idiomatic expressions can acquire the value of a cliché, as can any other linguistic item in a particular context. Makkai (1972) expresses a similar view in this respect: “Some clichés are idioms and some idioms are clichés, but neither group includes the other fully”.

Like idioms, euphemisms represent still another shifting pragmatic category that may turn into and function as clichés. An increasingly popular example is Sir Robert Armstrong’s famous statement during the cross-examination in which he tries to provide a justification for not having reported everything he knew in the Spy Catcher case. When asked why he did not tell the authorities everything he knew and withheld important information, his reply was: “I have been economical with the truth”. This example is symptomatic for the way in which many euphemisms are more politically biased nowadays. As a result of the growing influence of the concept of ‘political correctness’ in several areas of social and political life, an increasing number of ‘politically correct’ terms have emerged lately, some of them replacing older, no longer appropriate ones. Thus, the former ‘Swedish Board of Immigration’ has recently been renamed with a strategically more fitting name, i.e. the ‘Swedish Board of Integration’, which is generally perceived as an attempt to avoid the overuse of the negatively loaded word ‘immigration’.
Are there clichés specific to particular professions? On the one hand, there is a ‘jargonisation’ of all-purpose clichés, such as ‘deliver the goods’, which is increasingly used institutionally, on the other, there is an emergence of everyday clichés adapted from established institutional discourse types, such as ‘to know the ropes’, or ‘cast the anchor’, borrowed from seafaring jargon. Howard refers to these as ‘occupational clichés’ (1986: 90), while Ichheiser (1949) calls them ‘occupational stereotypes’ (1948: 33). Nowadays clichés are more widespread since they are used not only by professionals at the work place, but also by laymen outside the work place. More and more people are using, as well as misusing, clichés coined within various types of institutional discourse, such as journalese, advertising, legalese, political or medical jargon. This is why we propose to call them institutional clichés. The dynamics of institutional discourse is thus speeded up and this may account for the fact that some clichés become outdated and are gradually replaced with new ones.
Propaganda, another of those tired expressions one seldom hears nowadays because public relations has taken its place (or education, or consciousness-raising): the propagandist does not coax, wheedle, indoctrinate, or inveigle the public into accepting his point of view, but educates it or raises its consciousness. (Bolinger, 1980: 115)
Worth mentioning among institutional clichés are the ones originating in the theories of language and philosophy of language, such as ‘Colourless green ideas sleep furiously’, ‘The king of France is bald’, ‘Have you stopped beating your wife?’. These clichés are not simply recycled, but also reinterpreted, recontextualised and rediscussed both within and outside their institutional frame.
Gender clichés are age-old, but they have become so downgraded and deconstructed lately that they have already started to trigger their counter-clichés. For example, the masculine 3rd person pronoun ‘he’ used to act as the generic pronoun for both ‘he’ and ‘she’, whereas now it is normally used together with ‘she’, which in writing appears as ‘s/he’. Several writers have even made a point of consistently using ‘she’ instead of ‘he’ as a generic pronoun in their books. Moreover, many generic nouns ending in ‘man’, such as ‘spokesman’, ‘chairman’, have undergone an alteration, whereby ‘man’ has been replaced by ‘person’, as in ‘spokesperson’ and ‘chairperson’, the new clichés in the making.

An important category of clichés which are being institutionalised as often as they are being deinstitutionalised are ethnic clichés. They are as necessary as they can become infinitely dangerous in that they have a tendency to degenerate easily from outlines of recognizable national patterns of behaviour and mentality into polarised, usually negative and distorted, value judgements. Past and modern history shows clearly that the misuse of such ethnic clichés can have most undesirable, even tragic, effects. Take a common cliché, like ‘Swedes are blond and blue-eyed’. It probably started from a matter of fact observation, which may afterwards lend itself to a variety of more or less biased interpretations. According to one interpretation, which exists in Swedish, ‘blue-eyed’ also means ‘innocent, not sophisticated, easy to be cheated’. An oversimplification and generalisation of this particular connotation would not only be false, but also misleading.
Are clichés specific to a culture, and to a certain age? Certainly every culture and every age have their own clichés, but there are also commonly shared clichés. While it is obvious that culturespecific and time-specific clichés may cause certain problems of understanding and interpretation, it is less obvious that each culture and each historical age is differently aware of the commonly shared clichés. For example, in our age, postmodernist clichés in the visual arts and popular culture represent meaningful and easily recognizable expressions of thinking patterns and values. Many more or less universally recognized clichés seem to have fostered their own ‘subclichés’ in different ages and communities. For instance, a cliché like “man is the measure of all things” has been frequently reinterpreted. Today, as we know, the meaning of ‘man’ can equally be interpreted less generically and more in terms of the contrast between the two poles of the dichotomy ‘man’ – ‘woman’.

4. Clichés Revisited – Redefining Clichés as Discursive and Rhetorical Structures
As we have showed above, clichés are not definable in strictly linguistic terms. Like other complex elements of discourse, clichés have a multi-faceted structure: textual, ideational and pragmatic (Halliday 1989/85). When faced with a cliché we can notice one of these aspects, or two, or all of them. We may conceive of a cliché as a phrase, or a fixed expression (textual level). Or we may look upon a cliché as a certain idea, the propositional content of the particular phrase just mentioned or as a logical scheme underlying a number of stereotypical phrases as a generative matrix (ideational level). Or it may bring to mind a repetitive utterance or speech act (pragmatic level). Quite often all three aspects are involved in our evaluation of a cliché.
Can we find a differentia specifica, something that distinguishes clichés from other utteraces and ideas? In search of an answer to this question, we need to go back to Aristotle (Ross: 1949) and a very important distinction he made concerning two kinds of knowledge: apodictic and problematic. By using this distinction in our definition of clichés, we intend to show their particularity and relevance to argumentation.
To Aristotle, apodictic knowledge is certain and conclusively proved. It cannot be doubted. Its primary domain is mathematics and logic. It is the goal of a theoretical science. Problematic knowledge on the other hand concerns our practical life. It is relevant to situations where we must choose a line of action. In such cases we need to make a judicious decision, perhaps by arguing back and forth, but no certitude can be reached.
By applying both of these concepts we propose a rhetorical definition of a cliché in keeping with the functional approach outlined in section 3. above:
A cliché is a problematic claim treated like an apodictic truth by a certain group in a certain socio-cultural setting. Typically, it derives its discursive efficiency from a simple structure with general applicability. As can be seen, this definition relies primarily on the ideational aspect but does not exclude the other dimensions. The reason for this choice of focus is its emphasis on the functions of clichés in argumentative reasoning. This definition opens up opposite perspectives on clichés at the same time, thus doing justice to the somewhat dualistic nature of this concept.
Let us now see what the definition above actually entails. It is based on antagonistic concepts which help explain the distinctive functions of clichés in argumentation. What we find is a series of four dichotomies:
1. questionable vs unquestionable
On the one hand, a cliché expresses something problematic in the sense that it could actually be otherwise. It is not merely a formal truth of a logical nature. It makes a statement about the world, and in so doing it restricts the possible ways of looking at a certain issue. This is how a cliché becomes such a powerful tool for establishing and maintaining a common perspective upon reality. On the other, this problematic aspect tends to be forgotten when the cliché is used. Typically, a cliché is not questioned by its adherents. That does not mean that they might not see its problematic nature if it were pointed out to them. But basically they tend to treat the cliché as something that can be taken for granted and that is not a proper subject for a debate. This dichotomy can be further clarified if we consider a very well-known cliché: ‘All men are equal’. It sounds uncontroversially true, especially because it expresses something highly desirable. However, on closer examination, it becomes problematic because of the underlying questions in search of evidence:

How do we agree on what ‘equal’ means? Is it a measurable concept?, Does it apply in all circumstances?, etc.
2. implicit vs explicit
Since a cliché sounds so familiar, it may look acceptable and unproblematic to most people. Its apparent simplicity makes it easy to recall and gives it a special openness, like in ‘All men are equal’.
However, its ironical counter-cliché comes to mind instantly, to remind us that everything can be implicitly challenged: ‘Some men are more equal than others’. The initial cliché is meant to adjust and ‘correct’ precisely this circulating cynical view. Its explicit message is backed by the implicit assumption that there are exceptions to all generally accepted rules.

3. outsider’s perspective vs insider’s perspective
On the one hand, this dichotomy endorses the outsider’s view as an unbiased and critical perspective which makes it possible to see clichés for ‘what they are’, namely more or less challenging propositions that can always be questioned. On the other hand, it endorses the insider’s perspective, which makes it possible to look at clichés with the trusting eyes of the one who sees them as established and reliable points of reference. To take the example above, the use of the concept ‘equal’ may vary from one discourse community to another and from one historical age to another. What is ‘equality’ to some may very well be meaningless or downright unacceptable to others.

4. argumentative vs deductive
This dichotomy refers to the functioning of clichés in argumentation. To understand this, we have to see how a cliché is integrated into an argumentatively backed knowledge claim. On the one hand, argumentation deals with problematic issues, where deductive certainty cannot be achieved, and the questionable character of clichés allows for a considerable openness. On the other hand, an argumentation sets out to convince, to reduce uncertainty and ideally to replace it with assurance. The apparent self-evidence of clichés satisfies this striving for a firm ground on which to build a stable line of reasoning.
Let us now look at the further specifications of the definition of a cliché. First, we say that a typical cliché has a simple structure and is of a general nature. It does not apply primarily to a single individual or event but to all items of a kind. This wide applicability of a cliché is important to its use in argumentation. Having a general scope, it can serve as the major premise of a syllogistic type of argument. Or, in Toulmin’s terms (1958), it can function as a warrant, relating a ground to some specific claim. Usually, a warrant should not in itself be the object of a debate. The quasi-apodictic quality of the cliché makes it well suited for its supportive function. Second, a cliché is valid only within a specific group. This feature is highly relevant to argumentation. It helps explain why an argument must be directed at a certain audience, whether this audience consists of some physically present persons or of an abstracted and idealized group operating only in the speaker’s mind (cf Perelman’s “universal audience”, 1969). The argument is successful only as long as it takes into consideration what this group can agree upon as reasonable starting points for a line of reasoning. This includes group-specific clichés.
Clichés have a striking resemblance to the concept of topos in classical rhetoric. Topos is often translated into English by commonplace, and although this translation does not render the whole meaning of the Greek word, it matches the aspect of its meaning which comes closest to a cliché – a phrase or an idea known and accepted by many which could be put to good use in persuasion or, heuristically, taken as a starting point in the search for a specific truth in a concrete case.
In classical rhetoric, a distinction was made between general and particular topics, the former belonging to all kinds of discourse and the latter to specific types of discourse. In the same way, a distinction could be made between all-purpose clichés, anchored in the everyday cultural patterns of a community, and institutional clichés, shared by the members of a professionally restricted group, such as lawyers and scientists observant of a certain rhetorical paradigm. Toulmin (1958), among others, sees argumentation as field-specific. These fields could be related to sets of particular clichés, which are recycled in different areas.

5. Your Clichés Tell Us Who You Are: Functions and Malfunctions of clichés in Argumentation
In this section, we will take a closer look at the functions of clichés in argumentation. We will start from the basic assumption that clichés bridge the gap between the problematic and the apodictic poles of an argument. We will also discuss some aspects of clichés that are likely to give rise to divergent interpretations in a particular situation and how that may affect their functions in argumentative discourse.
Let us start with a simple cliché that acts as a prop in a line of argumentation. A case in point is the following example. The political opposition in Sweden is known to continuously attack the social democratic government with arguments such as: ‘The present government should be voted out of power, since unemployment [which is now unusually high in Sweden] can only be fought by setting the market forces free’. This type of argument is based on two powerful clichés which, although not self-evident in the strict sense, are not likely to be critically examined by supporters, or by unmotivated addressees. The first cliché functions as the suppressed major premise: ‘Unemployment is bad and should be fought’. Probably, one could envisage a coherent argument against this view. But it still remains a cliché because most people who share this view do not feel that it should be interpreted as a controversial view that needs to be defended. They rest assured that all sensible people will agree. Actually, most people who think like this might not even have the impression that they are committed to a specific view but rather that they are just expressing an objective fact of life.
There is one more hidden cliché in the argument above, which functions as the suppressed minor premise. It could be expressed as follows: ‘Social democrats indulge in thwarting the market’. This is a cliché of a different kind. Many people who hold this view are definitely aware that it is not a neutral, but a politically loaded truth and that others may completely disagree. Still, it is regarded by many as a sort of axiom, a political fact that need not be discussed and that should be readily accepted by any unbiased observer of the political scene. And finally, let us examine the argument itself, ‘The present government should be voted out of power, since unemployment can only be fought by setting the market forces free’. This statement openly expresses the conclusion, the only explicit part of the argument.

In such a case, two important functions are associated with the cliché:
(a) it fulfils an ideological function by defining certain basic views about potentially controversial issues as being most accurate and relevant;
(b) it functions as a device for strengthening group cohesion: ‘Show me what your clichés are, and I will tell you what company you keep’. We will return to these points shortly.

Clichés are of crucial importance to rhetorical argumentation for one simple reason: Argumentation must always start somewhere, and preferably with something generally accepted within its target group. The least controversial things are the commonly shared views that we take for granted without any further critical thinking. Views that are constantly repeated in a way not inviting discussion are most likely to function as the shared axiomatic wisdom of a discourse community. Clichés can function as premises in argumentation, i.e. as agreed-upon beliefs in debatable issues, because they are normally accepted as suitable starting points for a line of reasoning. They may sometimes be implicit, as the suppressed premises above, since they are so well-known that they can easily be inferred by the listeners or readers. Actually, asserting  them instead of presupposing them might even draw undue attention to them and consequently involve them in a debate as elements that can be scrutinized and questioned. Thus, the truncated syllogisms called enthymemes and considered by Aristotle as a characteristic of rhetoric, are typically conveyed by clichés. By being left unverbalised, the cliché must be supplied by the addressee. Certainly, that leaves plenty of room for indeterminacy – anyone can distort a cliché so as to fit their goal.

6. The Characteristic Features of a Cliché as a Source of Miscommunication
We will now discuss what we consider to be five major features of clichés. The intrinsic complexity and the complementarity of these features enables the clichés to function argumentatively in a predictable and efficient way. However, the more their efficiency is taken for granted, the more their conventional interpretation tends to be abandoned when it interferes with new context-sensitive reinterpretations, which are likely to bring about different effects perceptions.
1. Typical for a cliché is that it is recycled in discourse. As a matter of fact, this is how it becomes a cliché. It is used over and over again, explicitly or implicitly, to convey a socially accepted common ground or shared belief among the people discussing a particular topic within a particular field. In this way, a cliché acquires its quasi-apodictic character, which may sometimes be reduced to what is politically correct, serving as a constraint for so-called acceptable ways of reasoning.
Moreover, a circulating cliché tends to develop semantic ‘density’. It becomes permeated with additional connotations and acquires a multiplicity of meanings, which allow for wider acceptability. Sometimes, however, a recycled cliché may result in combinations of incompatible elements. Building an argument from clichés does not mean so much starting with a well-defined common ground as bridging two distinct positions by means of a comprehensive and flexible approach. The argumentative movement towards a common conclusion could be suitably illustrated by a triangle pointing upwards and resting on the cliché as its base, which represents the diversity of opinions brought together by a common discursive practice.
2. Since they are the product of constant repetition, clichés are typically expressed by formulaic expressions. In certain situations clichés need to be explicitily mentioned or at least evoked, in order to be transmitted to members of the discourse group who are not yet fully socialised into it, to remind violators of decorum of the basic premises still in force or, simply as a way of reasserting them. Particularly for the last purpose, formulaic expressions can be seen to facilitate the articulation of a coherent line of argumentation. At the same time, their ready-made form makes them heavily dependent on context, thus allowing for a variety of interpretations.
In a dialectical discussion, however, these expressions need not function as catch-phrases which immediately support a claim, but as starters, feelers or stepping stones leading to more specific ideas developed during the ensuing dialogue. Clichés may be eventually reconsidered, but by then they will have already served their function as ready-made tools for opening up a new area to reflection.
3. Clichés are also important devices for group cohesion. Since clichés are socially rooted, they tend to present reality as reflected in a collective practice, by pointing back to the group(s) and to the the ideas that fostered them. More than any other element of social cognition, clichés discriminate between different groups, while at the same time serving as a means of identification for the members of each group. They can also fulfil a positive role by creating a greater awareness among the members of a particular group about the normality and acceptability of unfamiliar or unusual beliefs and customs observed in other communities. In this way, clichés help to shape a mentally and culturally coherent audience that can be collectively affected by the socially inclusive appeal of an ethical argument. Group cohesion can be established intraculturally, i.e. within and between ethnic groups or individuals belonging to the same culture or discourse community, but also interculturally, i.e. between ethnic groups or individuals belonging to different cultures.
4. Clichés tend to be ideologically loaded. Since clichés serve as rhetorical devices for orienting the members of a group in the social environment, they systematically influence people’s beliefs, value judgements and actions. In doing so, they fulfil an ideological function. This function becomes even more prominent in situations where political power is exerted by one group against another. The ideological function of clichés is important for rooting arguments in common ground, in order to guarantee their function as supporters of the general social claims articulated by the group using them. Thus, clichés help maintain a common perspective which is essential for efficient argumentation.
While normally functioning as general matrices of meaning that other utterances adjust to, clichés also allow for contextualised meanings. This is why it is essential to be familiar with the various socio-political configurations of a particular culture, both synchronically and diachronically. Ignoring the clichés of certain areas of one’s culture or of another culture deprives us of a crucial means for accessing discourse meaning and intentionality.
5. An interesting feature of clichés is that they tend to attract counter-clichés. As solutions to open-ended problems, clichés are not complete in themselves because they automatically trigger complementary alternatives. Between them, a cliché and a counter-cliché tend to structure an argumentative dialogue and give it pluridimensional orientation, dynamic intentionality and a higher potential for truthfulness. Politically left-wing and right-wing clichés may contradict each other, but they are also interrelated through a pattern of left-right polarisation.

7. Clichés in a Nutshell – Conclusions
The aim of this paper has been to examine and redefine clichés in an attempt to identify their argumentative functions, as well as their liability for miscommunication. We claim that clichés are crucial to argumentative discourse, and that their potential for building arguments is closely linked to their tendency to trigger divergent interpretations in certain contexts.
When redefining clichés, we argue that they do not represent a linguistic category, but rather a pragmatic and a rhetorical category, and we emphasize their dualistic nature. As a rule, a typical cliché has a simple structure with general applicability, which may account for its use and misuse in argumentation. A cliché may serve as the explicit/implicit premise, or as the explicit/implicit conclusion, in a syllogistic argument.
We started from the basic assumption that clichés bridge the gap between the problematic and the apodictic poles of an argument, which helps to explain why they tend to attract counterclichés. The analysis has led to the insight that building an argument from clichés does not mean so much starting with a well-defined common ground, as bridging two distinct positions by means of a comprehensive and flexible approach.
Due to the fact that clichés emphasize group cohesion and adherence to a particular ideology, they are often used as slogans which can give rise to conflicting interpretations when recycled by opposite socio-political groups.
We would like to conclude with what may very well turn out to be a cliché about clichés: we cannot do without them, but we had better watch out.

REFERENCES
Aristotle. (1949). Prior and Posterior Analysis. Translated by W.D. Ross. Oxford: Clarendon Press.
Bolinger, Dwight. (1980). Language: The Loaded Weapon. London: Longman.
Brook, G.L. (1981). Words in Everyday Life. London: Macmillan. The Cambridge International Dictionary of English. (1995). Ed. P. Procter, Cambridge University Press.
Gramley, Stephan, and Kurt-Michael Pátzold. (1992). A Survey of Modern English. London and New York: Routledge.
Halliday, M.A.K. and Hassan, R. (1989/1985). Language, Context and Text: Aspects of Language in a Social-Semiotic Perspective. Oxford: Oxford University Press.
Howard, Phillip. (1986/1984). The State of the Language. Harmondsworth: Penguin Books.
Ichheiser, Gustav. (1949). Misunderstandings in Human Relations: A Study in False Social Perception. Chicago, Illinois: The University of Chicago Press.
Lerner, L. D. (1956). Cliché and Commonplace. Essays in Criticism, VI (3) The Longman Webster English College Dictionary. (1985). 2 nd edn. Harlow: Longman.
Luelsdorff, P. (1981). What Is a Cliché? Trier: LAUT Papers, Series A, no. 85.
Makkai, Adam. (1972). Idiom Structure in English. The Hague: Mouton.
Missimer, Connie. (1991). Is Stereotype the Twentieth-Century Equivalent of Heresy’? In: F. H. van Eemeren et al. (eds). Proceedings of the Second International Conference on Argumentation. Amsterdam: Sic Sat.
The Oxford English Dictionary. (1989). 2 nd edn. Eds. J.A. Simpson and E.S.C. Weiner. Oxford: Clarendon Press.
Partridge, E. (1954). Dictionary of Clichés. London: Routledge & Kegan Paul.
Perelman, C. and Olbrechts-Tyteca, L. (1969). The New Rhetoric. Notre Dame: University of Notre Dame Press.
Redfern, Walter. (1989). Clichés and Coinages. Oxford: Basil Blackwell.
Ricks, C. (1980). Clichés. In: L. Michaels and C. Ricks (eds). The State of the Language. Berkeley: University of California Press, 54-63.
Toulmin, Stephen E. (1958). The Uses of Argument. London: Cambridge University Press.

1. The logic of law
Most lawyers have some awareness of logic, although the awareness is normally limited. The logical connectives ‘… and …’ and ‘ … or …’ are known, and maybe even the ambiguous interpretation of a composite sentence of the form ‘a and b or c’ is familiar. Some might regard the connective ‘if …, then …’ as the abstract form of a legal rule and the rule of inference Modus Ponens as the general template of legal reasoning.
Why do lawyers pay so little attention to logic? The main problem is that logic in its classical appearances (such as propositional or predicate logic) is not sufficiently satisfying as a model of legal argument: it is too far from the argument forms that lawyers use in practice. In recent years, there has been a large amount of research on the development of logical tools for legal argument (see, e.g., the work of Gordon [1993, 1995], Hage [1997], Lodder [1998], Prakken [1993, 1997] and Verheij [1996]). Argument forms that have been studied include arguments concerning exceptions to rules, conflicts of reasons and rule applicability.
The logical tools that have recently been developed can be categorized under three headings: defeasibility, integration of logical levels, and the process character of argument [Verheij et al., 1997]. Defeasibility is a characteristic of arguments and, in a derived sense, of conclusions. A conclusion is defeasible if it is the conclusion of a defeasible argument. Defeat occurs if a conclusion is no longer justified by an argument because of new information. For instance, the conclusion that a thief should be punished is no longer justified if it turns out that there was a legal justification for the theft, such as an authorized command.

The integration of logical levels is for instance required if reasons are weighed. If arguments lead to incompatible conclusions, weighing of reasons is necessary to determine which conclusion follows. Additional information is necessary to determine the outcome of the weighing process. In some views, this information is on a higher logical level than the facts of cases, and the rules of law. However, since there can also be arguments about the weighing of reasons, the integration of levels is required.
The process character of argument also led to the development of new logical tools. For instance, the defeasibility of arguments cannot be separated from the process of taking new information into account. During the process of argumentation conclusions are drawn, reasons are adduced, counterarguments are raised, and new premises are introduced. In traditional models, only the end products of the process are modeled.

The focus has been primarily on the technical development of the logical tools, and only in the second place on their practical adequacy for modeling legal argument. Presently a convergence of opinions on the necessary logical tools takes shape, and a systematic practical assessment of the logical tools becomes essential.  In the research reported on in this paper, a step towards the practical assessment is made by the development of two experimental computer systems for argument mediation for lawyers. In computer-supported argument mediation, one or more users of the system engage in an argument that is mediated by the system: the system administers the argument moves and safeguards that the rules of argument are observed. It can, if appropriate, give advice to the user.
A new problem for argument researchers, as posed by the development of systems for argument mediation is how arguments should be presented to the users of the system. In this paper, we describe two experimental computer systems, the Argue!-system and the Argumentation Mediator, each using a different way of argument presentation. The two systems are based on a simplified version of Verheij’s [1996] CumulA-model, which is a procedural model of argumentation with arguments and counterarguments.

Section 2 briefly discusses argument mediation and the two experimental systems of the present paper. In section 3, an example case of Dutch tort law is summarized, that will be used to illustrate the two systems of argument mediation. Section 4 contains an introduction of CumulA, the procedural model of argumentation with arguments and counterarguments, that underlies the two experimental systems. Section 5 and 6 contain sample sessions of the two systems. In section 7, the two systems are compared with each other and selected related systems, especially with regards to their underlying argumentation theories and user interfaces. Section 8 suggests a shift from argument mediation systems as theoretical to practical tools.[i]

2. Experiments with argument mediation
In the research on computer-mediated legal argument, computer systems are developed that can be used to mediate the process of argumentation of one or more users. The systems can mediate the process in which arguments are drafted and generated by the users, e.g., by
– administering and supervising the argument process,
– keeping track of the conclusions that are justified, and the assumptions that are made,
– keeping track of the reasons adduced and the conclusions drawn,
– keeping track of the counterarguments that have been adduced,
and
– checking whether the users of the system obey the pertaining rules of argument.

Recently several experimental systems for (legal) argument mediation have been developed (e.g., Room 5 by Loui et al. [1997], Zeno by Gordon and Karacapilidis [1997], and DiaLaw by Lodder [1998]). The systems differ on the user interfaces and on the underlying argumentation theory that is used.[ii]
A new problem for argument researchers, as posed by the development of systems for argument mediation is how arguments should be presented to the users of the system. In this paper, we describe two experimental computer systems, using different ways of argument presentation. The first, the Argue!-system, has a graphical user interface: the user of the system ‘draws’ argument structures, by clicking and dragging a pointing device, such as a mouse. The second, the Argumentation Mediator, has a template-based user interface: the user gradually constructs arguments, by filling in templates that correspond to argument patterns.
The two systems are based on Verheij’s [1996] CumulA-model, which is a procedural model of argumentation with arguments and counterarguments.

3. An example taken from Dutch tort law: the ‘bussluis’ case
To illustrate the two systems of argument mediation, we use an example taken from Dutch tort law. In Dutch tort law, the liability to repair damages on the basis of tort is determined in two steps:

Step 1. Determine the general duty to compensate damages on the basis of tort (art. 6:162 BW)
Step 2. Determine the relative amount of imputability in order to find the portion of the damages that has to be compensated (art. 6:101 BW)

For instance, assume that John has the general duty to compensate for certain damages, as suffered by Mary, on the basis of a tort committed by John. Assume also that the damages were partly due to Mary’s own fault. If the judge decides that the damages must be imputed to Mary for 25 %, John only has to compensate 75 % of Mary’s damages.

The logic of Dutch tort law enabled the somewhat surprising decision in the Dutch ‘bussluis’ case between a cab-driver and the local authorities (Dutch Supreme Court, March 20, 1992; Court of Justice of the Hague, September 15, 1994): although there was a general duty of the Municipality to compensate for the damages of the cab-driver, the actual portion of the damages that had to be compensated for was nil, because the damages were fully imputed to the cab-driver.

The reasoning can be summarized as follows:
Step 1. The Municipality had committed a tort against the cab-driver.
Therefore, the Municipality had the general duty to repair the damages (on the basis of art. 6:162 BW).
Step 2. The damages were fully imputed to the cab-driver. Therefore, the portion of the damages to be compensated for was nil (on the basis of art. 6:101 BW).

The case is discussed more extensively by Lodder and Verheij [1998] and Verheij and Lodder [1998].

4. CumulA: a model of defeasible argumentation in stages
CumulA [Verheij, 1996] is a procedural model of argumentation with arguments and counterarguments. It is based on two main assumptions. The first assumption is that argumentation is a process during which arguments are constructed and counterarguments are adduced. The second assumption is that the arguments used in argumentation are defeasible, in the sense that whether they justify their conclusion depends on the counterarguments available at a stage of the argumentation process.

The goal of argumentation is to (rationally) justify conclusions. In CumulA, the focus is on the process of argumentation, and on the defeasibility of the arguments used in argumentation. Argumentation is a process, in the sense that during argumentation arguments are constructed and counterarguments are brought up. Arguments are assumed to be defeasible, in the sense that if an argument at some stage of the argumentation process justifies its conclusion, it not necessarily justifies its conclusion at all later stages. The defeat of an argument is caused by a counterargument that is itself undefeated.

For instance, if the Municipality has committed a tort against the cab-driver, a conclusion would be that the Municipality has the duty to repair 100 % of the damages. The conclusion can be rationally justified, by giving support for it. E.g., the following argument could be given:

The Municipality has committed a tort against the cab-driver.
So, the Municipality has the (general) duty to repair the damages.
So, the Municipality has the duty to repair 100 % of the damages.

Recall that in Dutch tort law, the general duty to repair damages and the portion of the damages to be repaired are established consecutively.

An argument as above is a reconstruction of how a conclusion can be supported. The argument given here consists of two steps.
An argument that supports its conclusion does not always justify it. For instance, if in our example it turns out that the damages are fully imputed to the cab-driver (as in the ‘bussluis’ case), the conclusion that the Municipality has the duty to repair 100 % of the damages would no longer be justified. The argument has become defeated. In the example, the argument:
The Municipality has the (general) duty to repair the damages.
So, the Municipality has the duty to repair 100 % of the damages.

does not justify its conclusion because of the counterargument.
The damages are fully imputed to the cab-driver.

CumulA is a procedural model of argumentation with arguments and counterarguments, in which the defeat status of an argument, either undefeated or defeated, depends on:
(1) the structure of the argument;
(2) the counterarguments;
(3) the argumentation stage.

We briefly discuss each below. The model builds on the work of Pollock [1987, 1995], Loui [1991, 1992], Vreeswijk [1993, 1997] and Dung [1995] in philosophy and artificial intelligence, and was developed to complement the work on the model of rules and reasons Reason-Based Logic (see, e.g., Hage [1993, 1996, 1997] and Verheij [1996]).

In the model, the structure of an argument is represented as in the argumentation theory of Van Eemeren and Grootendorst [1981, 1987]. Both the subordination and the coordination of arguments are possible. It is explored how the structure of arguments can lead to their defeat. For instance, the intuitions that it is easier to defeat an argument if it contains a longer chain of defeasible steps (‘sequential weakening’), and that it is harder to defeat an argument if it contains more reasons to support its conclusion (‘parallel strengthening’), are investigated.
In the model, which arguments are counterarguments for other arguments is taken as a primitive notion [cf. Dung, 1995]. This is in contrast with Vreeswijk’s [1993, 1997] model, in which conflicts of arguments (i.e., arguments with conflicting conclusions) are the primitive notion. In CumulA, so-called defeaters indicate which arguments are counterarguments to other arguments, i.e., which arguments can defeat other arguments. It turns out that defeaters can be used to represent a wide range of types of defeat, as proposed in the literature, e.g., Pollock’s [1987] undercutting and rebutting defeat.
Moreover some new types of defeat can be distinguished, namely defeat by sequential weakening (related to the well-known sorites paradox) and defeat by parallel strengthening (related to the accrual of reasons).
In the CumulA-model, argumentation stages represent the arguments and the counterarguments currently taken into account, and the status of these arguments, either defeated or undefeated. The model’s lines of argumentation, i.e., sequences of stages, give insight in the influence that the process of taking arguments into account has on the status of arguments.
For instance, assume that John has the general duty to compensate for certain damages, as suffered by Mary, on the basis of a tort committed by John. Assume also that the damages were partly due to Mary’s own fault. If the judge decides that the damages must be imputed to Mary for 25 %, John only has to compensate 75 % of Mary’s damages.

To summarize, CumulA shows
(1) how the subordination and coordination of arguments is related to their defeat;
(2) how the defeat of arguments can be described in terms of their structure, counterarguments, and the stage of the argumentation process;
(3) how both forward and backward argumentation can be formalized in one model.

Verheij [1996] discusses the CumulA-model extensively, both informally and formally. The two argument mediation systems, discussed in this paper, have restricted versions of CumulA as their underlying argumentation theory.

5. The Argue!-system: a graphical interface
The first experimental implementation of CumulA is an argument mediation system with a graphical interface. It is referred to as the Argue!-system. The user ‘draws’ argument structures, by clicking and dragging a pointing device, such as a mouse. We discuss an example session, based on the ‘bussluis’ case. As a start, a statement is typed, ‘The Municipality has committed a tort against the cab-driver’ (Figure 1).

Statements can be justified by adding reasons (in the figure: ‘The Municipality has acted against proper social conduct’), and can be used to draw conclusions (‘The Municipality has the duty to repair the damages’). This is visually depicted in a straightforward way, by arrows connecting the statement-boxes (Figure 2).

The reader may have noticed that the statement ‘The Municipality has committed a tort against the cab-driver’ was first in a grey box, and now is in a white box. This is due to the different statuses that statements can have: if a statement is unevaluated it is in a grey box, if it is undefeated (i.e., justified), it is in a white box. In the example, the statement ‘The Municipality has acted against proper social conduct’ is undefeated, since it has been added as an assumption. The other two statements become undefeated since there is an undefeated reason for them.

The line of argument continues in order to determine the amount of damages that the Municipality has to pay. At first, the conclusion is drawn that the Municipality has the duty to repair 100 % of the damages. However, the user recalls something about the importance of imputability (Figure 3).

The statement that the damages are fully imputed to the cab-driver is a counterargument to the argument that the Municipality has the duty to repair 100 % of the damages because the Municipality has committed a tort against the cab-driver. In order to indicate that one argument is a counterargument to another, a special visual structure is used (Figure 4).

Since the statement that the damages are fully imputed to the cab-driver is as yet unevaluated, the statement that the Municipality has the duty to repair 100 % of the damages is still justified. In order to justify the statement that the damages are fully imputed to the cab-driver, the relevant case is cited. Since the corresponding statement that the Court decided on the imputability, is added as a assumption, the conclusion that the damages are fully imputed to the cab-driver, becomes justified (Figure 5).

As a side effect, the statement that the Municipality has the duty to repair 100 % of the damages, has become defeated (visually indicated by the cross in the corresponding box), since the argument that the damages are fully imputed to the cab-driver, now is a counterargument.

Now it is concluded that the Municipality has the duty to repair 0 % of the damages, on the basis of the reason that the damages are fully imputed to the cab-driver. If desired, the rule that warranted the connection between the reason and the conclusion, can be made explicit by the user of the system (Figure 6).

When the user has stated that the rule of art. 6:102 of the civil code determines the portion of the damages, the session ends (Figure 7).

6. The Argumentation Mediator: a template-based interface

The second experimental implementation of CumulA is a system for the mediation of argument with a template-based interface. It is referred to as the Argumentation Mediator. The user gradually constructs arguments, by filling in templates that correspond to argument patterns. We give an example session, again based on the ‘bussluis’ case. The opening screen of the implementation shows four ‘Argue’-buttons[iii], that give access to the available argument templates, and four ‘View’-buttons, that give different ways of viewing the current stage of argumentation stage (Figure 8). In the example session, the functionality of the buttons will be explained.

When the user clicks the ‘Statement’-button, a template for making a statement is shown (Figure 9). The user types the statement that the Municipality has committed a tort against the cab-driver. The statement can have be of two types, namely the query- and the assumption-type. A statement of query-type is a new issue[iv] of argumentation and no claim is made with regards to its justification status. Normally the goal of making a query-type statement is to determine whether it is justified. A statement of assumption-type is taken as justified ‘by assumption’ and does not require further justification. Normally the goal of making an assumption-type statement is to use it for justifying other statements (of query-type). In the example, the statement that the Municipality has committed a tort against the cab-driver, is of query-type, since the user wants to establish whether it is justified.

The result of the argument move is the following. The icon in front of the sentence ‘The Municipality has committed a tort against the cab-driver’ consists of a question mark, indicating that the corresponding statement is of query-type, and a (grey) circle, indicating that it is currently neither justified nor unjustified (Figure 10). Now the user clicks the ‘Reason/conclusion’-button to give a reason for the conclusion that the Municipality has committed a tort against the cab-driver (Figure 11).

Since reason that the Municipality has acted against proper social conduct is of assumption-type (indicated by the exclamation mark in its icon), the conclusion that the Municipality has committed a tort becomes justified, indicated by the (green) plus. Since the reason is of assumption-type it is also taken as justified (Figure 12).

The ‘Reason/conclusion’-template can of course not only be used for adducing reasons, but also for drawing conclusions. Below the user uses the statement that the Municipality has committed a tort as a reason to draw the conclusion that the Municipality has the duty to repair the damages (Figure 13).

From the reason that the Municipality has the duty to repair the damages the user draws the conclusion that the Municipality has the duty to repair 100 % of the damages (Figure 14).

The user recalls that the imputability of the cab-driver can diminish the amount of damages to be repaired. By clicking the ‘Exception’-button the user gets access to the exception-template. The user types the exception that the damages are fully imputed to the cab-driver, and selects the reason/conclusion it blocks (Figure 15).[v]

The result of the user’s exception move is that the conclusion that the Municipality has the duty to repair 100 % of the damages is no longer justified, as indicated by the (red) cross (Figure 16).

Although the exception that the damages are fully imputed to the cab-driver was an assumption, the user chooses to give a reason for it (using the reason/conclusion-template), namely that the Court decided on the imputability (Sept. 15, 1994). The type of the exception is changed to the query-type (Figure 17)

Finally, the user concludes that the Municipality has the duty to repair 0 % of the damages (Figure 18).

Until now, the session always showed the arguments that were constructed, since the ‘Arguments’-button of the ‘View’-panel was pressed. The other buttons of that panel give access to other information. For instance, the ‘Line of argumentation’-button gives access to the successive argument moves performed by the user (Figure 19).

The ‘Statements’- and ‘Reasons’-buttons give access to the statements and the reasons (including the corresponding conclusions and exceptions) that have been entered by the user (Figure 20, 21).

7. A comparison of argument-mediation systems

In order to put the two discussed systems for argument mediation in context, they are briefly compared to three other systems, namely Room 5 by Loui et al. [1997], Zeno by Gordon and Karacapilidis [1997], and DiaLaw by Lodder [1998]. The system of section 5 is referred to as the Argue!-system, that of section 6 as the Argumentation Mediator. First, the underlying argumentation theories are discussed; second, the user interfaces.

7.1 The underlying argumentation theories
In the underlying argumentation theories of all five systems argumentation is dynamic. Statements can be made, and reasons can be adduced. In Room 5, Zeno and DiaLaw, argumentation is issue-based. No new conclusions can be drawn, since these systems focus on justification of an initial central issue. In the Argue!-system (section 5) and the Argumentation Mediator (section 6), argumentation is free, since there is no central issue, and allow both forward argumentation (i.e., drawing conclusions) and backward argumentation (i.e., adducing reasons).

All systems model a notion of defeasibility of argumentation. Room 5, Zeno and DiaLaw have a notion of reasons for and against conclusions.[vi] In Zeno and DiaLaw, weighing the conflicting reasons determines which conclusions are justified. DiaLaw, the Argue!-system and the Argumentation Mediator have an undercutter-type exception (see note 5). The Argue!-system models defeat by sequential weakening (see Verheij [1996, p. 122]): an argument is defeated since it contains an unacceptable sequence of steps.

Only DiaLaw has a notion of the rules underlying argument steps, as it is based on the theory of rules and reasons Reason-Based Logic (see, e.g., Hage [1996, 1997] and Verheij [1996]).

In Room 5, Zeno and DiaLaw, argumentation is considered as a game with participants. In Room 5 and Zeno, the game character is left implicit, but obtained by the distributed access to the systems, on the World-Wide Web. In DiaLaw, the game character is made explicit in the form of a dialogue game with two parties. The Argue!-system and the Argumentation Mediator have no explicit notion of participants.

7.2 The user interfaces
Room 5, Zeno, the Argue!-system and the Argumentation Mediator present arguments in a visual manner. Zeno, the Argue!-system and the Argumentation Mediator use a tree-like presentation. Room 5 uses a clever system of boxes-in-boxes in an attempt to avoid ‘pointer-spaghetti’. In DiaLaw, and also in the Argumentation Mediator, argumentation is presented in a verbal manner, namely as a sequence of moves.

In Room 5 and Zeno, counterarguments (formed by reasons against conclusions) are grouped together in the visual argument structure. In the Argue!-system, counterarguments are shown by a special visual structure. In the Argumentation Mediator, counterarguments (currently only formed by undercutter-type exceptions) are visible in a special viewing window, namely the ‘Reasons’-view. In DiaLaw, counterarguments are not directly accessible.

In the Argumentation Mediator and DiaLaw, the dynamic aspect of argumentation is shown by a view on the sequence of moves. In Room 5, Zeno and the Argue!-system, only a view on the current stage of the argumentation process is visible. In Room 5 and the Argumentation Mediator, it is possible to switch between different views showing different types of information.

DiaLaw has a text-based interface; moves are typed at a command-prompt. The Argue!-system has a graphical interface: argument structures are ‘drawn’ using a pointing device. Room 5, Zeno and the Argumentation Mediator have a template-based interface: users fill in forms to perform an argument move. The Argumentation Mediator provides different forms for different moves to facilitate the user.

7.3 Conclusions
If we look at the above discussion, some conclusions can be drawn.
– An issue-based argument mediation system has the advantage that the process of argumentation has a focus, which can be useful, or even necessary (e.g., in a game-like situation). However, a system that is not issue-based (such as the two systems presented in this paper) adds flexibility, namely the possibility of forward argumentation.
– Current argument mediation systems have different notions of defeasibility. One should therefore strive for integration, or explicitly defend choices.
– A notion of rules should be included, or one should defend why it is not. Remarkably, none of the discussed systems with a visual, window-style interface has a notion of rules.
– Argument mediation systems with visual, window-style interfaces are obviously more user-friendly than text-based interfaces.

Among the visual interfaces, a template-based interface seems easier to use than a graphical interface (as in the Argue!-system), in which special visual structures have to be drawn. A system with different templates for different types of argument moves (as in the Argument Mediator) seems promising.

– The choice of argument moves that are available to the user, is crucial for user acceptance. A particular choice should not just be based on theoretical grounds, but must correspond to the needs of users.

8. Argument mediation systems for lawyers: from theoretical to practical tools
The recent advances in the theory of legal argument, especially with respect to defeasibility, integration of logical levels, and the process character of argument require a practical assessment. One way of such assessment is to build usable systems for argument mediation. In this paper, two experimental systems, namely the Argue!-system and the Argumentation Mediator, have been presented, and briefly compared to selected related systems, namely Room 5, Zeno, and DiaLaw. The differences between the underlying argumentation theories and user interfaces are striking, and show that argument mediation systems are still in their early stages of development. On the one hand, current argument mediation systems seem not yet sufficiently mature to be used as practical tools by practicing lawyers. On the other hand, they already turn out useful as theoretical tools, and help to enhance argumentation theory. The move from theoretical to practical tools will take serious effort, both by researchers and by system developers, but is manageable for the near future.

Acknowledgments
The author gladly acknowledges the financial support by the Dutch National Programme Information Technology and Law (ITeR) for the research reported in this paper (project number 01437112). He also thanks Arno Lodder and Jaap Hage for comments and discussion.

NOTES
[i] Sections 4 and 5 have been adapted from Lodder and Verheij [1998] and Verheij and Lodder [1998].
[ii] In the argumentation theories of all systems for argument mediation discussed in this paper, argumentation is considered defeasible. Systems based on classical logic, e.g., Tarski‘s World by Barwise and Etchemendy
(see http://csli-www.stanford.edu/hp/) are not discussed.
[iii] At the time of writing this paper, the ‘Pros & cons’-button does not yet give access to a template.
[iv] This is the terminology used in the Zeno-project [Gordon and Karacapilidis, 1997].
[v] The exception is of undercutter-type [Pollock, 1987], as it breaks the justifying connection between the reason and the conclusion. I slightly prefer to speak of exceptions to rules, and not to reasons, that arise from rules (see the work on Reason-Based Logic by Hage [1996, 1997] and Verheij [1996]). However, the current implementation does not give access to the rules behind reasons.
[vi] The ‘Pros & cons’-button of the Argumentation Mediator suggests it has a notion of reasons for and against conclusions. However, as yet, no corresponding functionality has been implemented (see note 3).

REFERENCES
Dung, P.M. (1995). On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games. Artificial Intelligence, Vol. 77, pp. 321-357.
Eemeren, F.H. van, Grootendorst, R., and Kruiger, T. (1981). Argumentatietheorie. Uitgeverij Het Spectrum, Utrecht.
Eemeren, F.H. van, Grootendorst, R. and Kruiger, T. (1987). Handbook of Argumentation Theory. A Critical Survey of Classical Backgrounds and Modern Studies. Foris Publications, Dordrecht. Translation of van Eemeren et al. (1981).
Gordon, T.F. (1993). The Pleadings Game. An Artificial Intelligence Model of Procedural Justice. Dissertation.
Gordon, T.F. (1995). The Pleadings Game. An Artificial Intelligence Model of Procedural Justice. Kluwer Academic Publishers, Dordrecht.
Gordon, T.F., and Karacapilidis, N. (1997). The Zeno Argumentation Framework. The Sixth International Conference on Artificial Intelligence and Law. Proceedings of the Conference, pp. 10-18. ACM, New York (New York).
Hage, J. (1993). Monological reason based logic. A low level integration of rule-based reasoning and case-based reasoning. The Fourth International Conference on Artificial Intelligence and Law. Proceedings of the Conference, pp. 30-39. ACM, New York (New York). Also published as report SKBS/B3.A/93-08.
Hage, J. (1996). A Theory of Legal Reasoning and a Logic to Match. Artificial Intelligence and Law, Vol. 4, pp. 199-273.
Hage, J.C. (1997). Reasoning with Rules. An Essay on Legal Reasoning and Its Underlying Logic. Kluwer Academic Publishers, Dordrecht.
Lodder, A.R. (1998). DiaLaw – on legal justification and dialog games. Dissertation, Universiteit Maastricht.
Lodder, A.R., and Verheij, B. (1998). Opportunities of computer-mediated legal argument in education. Proceedings of the BILETA-conference – March 27-28. Dublin, Ireland.
Loui, R.P. (1991). Ampliative Inference, Computation, and Dialectic. Philosophy and AI. Essays at the Interface (eds. Robert Cummins and John Pollock), pp. 141-155. The MIT Press, Cambridge (Massachusetts).
Loui, R.P. (1992). Process and Policy: Resource-Bounded Non-Demonstrative Reasoning. Report WUCS-92-43. Washington University, Department of Computer Science, Saint Louis (Missouri).
Loui, R.P., Norman, J., Altepeter, J., Pinkard, D., Craven, D., Lindsay, J., and Foltz, M. (1997). Progress on Room 5. A Test-bed for Public Interactive Semi-Formal Legal Argumentation. The Sixth International Conference on Artificial Intelligence and Law. Proceedings of the Conference, pp. 207-214. ACM, New York (New York).
Pollock, J.L. (1987). Defeasible reasoning. Cognitive Science, Vol. 11, pp. 481-518.
Pollock, J.L. (1995). Cognitive Carpentry: A Blueprint for How to Build a Person. The MIT Press, Cambridge (Massachusetts).
Prakken, H. (1993). Logical tools for modelling legal argument. Doctoral thesis, Vrije Universiteit, Amsterdam.
Prakken, H. (1997). Logical Tools for Modelling Legal Argument. A Study of Defeasible Reasoning in Law. Kluwer Academic Publishers, Dordrecht.
Verheij, B. (1996). Rules, Reasons, Arguments. Formal studies of argumentation and defeat. Dissertation Universiteit Maastricht.
A summary and table of contents are available on the World-Wide Web at http://www.metajur.unimaas.nl/~bart/proefschrift/.
Verheij, B., Hage, J.C., and Lodder, A.R. (1997). Logical tools for legal argument: a practical assessment in the domain of tort. The Sixth International Conference on Artificial Intelligence and Law. Proceedings of the Conference, pp. 243-249. ACM, New York (New York). An abstract is available on the World-Wide Web at http://www.metajur.unimaas.nl/~bart/papers/icail97.htm.
Verheij, B., and Lodder, A.R. (1998). Computer-mediated legal argument: the verbal vs. the visual approach. Proceedings of the 2nd French-American Conference on AI and Law – June 11-12. Nice, France.
Vreeswijk, G.A.W. (1993). Studies in defeasible argumentation. Doctoral thesis, Vrije Universiteit, Amsterdam.
Vreeswijk, G.A.W. (1997). Abstract argumentation systems. Artificial Intelligence, Vol. 90, pp. 225-279.

Argumentative explicitness is commonly acknowledged to be a normative ideal for argumentative practice, but advocates might fear that explicit argumentation could impair persuasive success. The question of the persuasive effects of argumentative explicitness is an empirical one, however. This paper addresses one aspect of this matter, by offering a meta-analytic review of the persuasive effects associated with one aspect of the degree of articulation given to an advocate’s supporting argumentation, namely, whether the advocate explicitly identifies the sources of supporting information.

1. Background
Argumentative explicitness is one commonly-recognized normative good in the conduct of advocates. That is, it is normatively desirable that advocates explicitly articulate their viewpoints: “Evasion, concealment, and artful dodging . . . are and should be excluded from an ideal model of critical discussion” (van Eemeren, Grootendorst, Jackson, & Jacobs 1993: 173). Explicit argumentation is normatively desirable because explicitness opens the advocated view for critical scrutiny. But explicit argumentation might not be instrumentally successful, that is, persuasive, which gives rise to the question: what is the relationship between argumentative explicitness and persuasive effects?
One facet of this question has been addressed by O’Keefe (1997), who reviewed research concerning the persuasive effects of variations in the explicitness of a message’s conclusion (the degree of articulation of the message’s overall standpoint or recommendation). His review suggested that better-articulated message conclusions are dependably more persuasive than less-articulated ones.
This paper concerns the persuasive effects of variation in the explicitness of one facet of a message’s supporting argumentation, specifically, whether the advocate explicitly identifies the sources of provided information. A number of studies have addressed this question, though many of these have never been systematically collected or reviewed. The purpose of the present paper is to provide a meta-analytic review of this research.

Meta-analytic literature reviews aim at providing systematic quantitative summaries of research studies (Rosenthal 1991 provides a useful general discussion of meta-analysis). Traditional narrative literature reviews emphasize statistical significance (whether a given study finds a statistically significant effect), but this can be a misleading way of characterizing research findings; whether statistical significance is achieved is a matter of, inter alia, sample size. Meta-analytic reviews instead commonly focus on the size of the effect obtained in each study, with these then being combined to give an observed average effect (with an associated confidence interval). In this paper, the effect of central interest is the persuasive outcome associated with variation in information-source citation.
A number of studies relevant to this question are ones commonly characterized as studies of the effects of “evidence” in persuasive messages (e.g., McCroskey 1969; Reinard 1988). The question of interest in these studies is what difference it makes to persuasive effectiveness if the advocate provides evidence supporting the message’s claims. As Kellermann (1980) has pointed out, however, the concept of evidence invoked in this research is not carefully formulated and, correspondingly, evidence research has seen a large number of different experimental realizations of evidence variations (see Kellermann 1980: 163-164). Kellermann has argued quite pointedly for the importance of more careful conceptualization of the relevant message properties.
One of the message variations commonly represented in evidence research is information-source citation. That is, as part of manipulating the presence of “evidence” in a message, investigators have varied whether the message contains explicit identification of information sources. Thus in a number of studies, information-source citation has been manipulated simultaneously (that is, in a confounded fashion) with other variables (e.g., Harte 1972; McCroskey 1966).
The present review thus has a somewhat sharper focus than those in discussions of evidence, by virtue of being concerned specifically with information-source citation (cf., e.g., Reinard 1994). This more careful specification of the message property of interest has also made it possible to locate relevant research not commonly mentioned in discussions of evidence (e.g., Berger 1988). Moreover, given that some studies have manipulated information-source citation in tandem with other variables, the present focus permits one to distinguish cases in which only information-source citation is varied from cases that simultaneously vary information-source citation and other message properties; studies of such joint manipulations are of distinctive interest, precisely because they shed light on the question of the effects of combining information-source citation manipulations with other variations.

2. Method
Identification of Relevant Investigations
Literature search. Relevant research reports were located through personal knowledge of the literature, examination of previous reviews and textbooks, and inspection of reference lists in previously-located reports. Additionally, searches were made through databases and document-retrieval services using such terms as “documentation,” “evidence,” and “support” in conjunction with “persuasion” and “persuasive” as search bases; these searches covered material through at least January 1998 in PsycINFO, ERIC (Educational Resources Information Center), Current Contents, ABI/Inform, and Dissertation Abstracts Online.
Inclusion criteria. Studies selected had to meet two criteria. First, the study had to compare two messages varying in information-source citation; specifically, the study had to contrast a message that explicitly identified the sources of (at least some of) the message’s information (facts, opinions, and the like) and a message that presented the same information without such identifying source information. This criterion excluded studies that varied other aspects of the message’s explicitness, such as the explicitness of the overall conclusion (e.g., Hovland & Mandell 1952), the completeness with which supporting-argument premises or conclusions were articulated (e.g., Kardes 1988), and the like.
Second, the investigation had to contain appropriate quantitative data pertinent to the comparison of persuasive effectiveness or perceived credibility between experimental conditions. This criterion excluded studies that did not provide appropriate quantitative information about effects (e.g., Babich 1971; Kilcrease 1977; McCroskey 1967b, studies 2, 6, 11, 12, and 13).

Dependent Variables and Effect Size Measure
Dependent variables. Two dependent variables were of interest. The dependent variable of central interest was persuasiveness (as assessed through measures such as opinion change, postcommunication agreement, behavioral intention, and the like). When a single study contained multiple indices of persuasion, these were averaged to yield a single summary.
The other dependent variable was credibility (as assessed though, e.g., measures of competence, trustworthiness, believability, and the like). Where multiple indices of credibility were available, these were averaged.

Effect size measure. Every comparison between a message providing information-source citations and its less explicit counterpart (without such citations) was summarized using r as the effect size measure. Differences favoring explicit messages were given a positive sign; differences favoring inexplicit messages were given a negative sign.
When correlations were averaged across several dependent measures, the average was computed using the r-to-z-to-r transformation procedure, weighted by n. Wherever possible, multiple-factor designs were analyzed by reconstituting the analysis such that individual-difference factors (but not other experimental manipulations) were put back into the error term (following the suggestion of Johnson 1989).
When a given investigation was reported in more than one outlet, it was treated as a single study and analyzed accordingly. The same research was reported (in whole or in part) in Cathcart (1953) and Cathcart (1955); in Harte (1972) and Harte (1976); in Hayes (1966) and Hayes (1971); in Luchok (1973) and in Luchok and McCroskey (1978), recorded here under the latter; in McCroskey (1967b, Study 1), McCroskey (1966, pilot study), McCroskey (1967a), and in McCroskey and Dunham (1966, Experiment 1); in McCroskey (1967b, Study 2) and in McCroskey and Dunham (1966, Experiment 2); in McCroskey (1967b, Study 3) and in Holtzman (1966); in McCroskey (1967b, Study 4) and in McCroskey (1966, major study I); in McCroskey (1967b, Study 5) and in McCroskey (1966, major study II); in Ostermeier (1966) and Ostermeier (1967); in Reinard (1984, Experiment 1) and in Reinard and Reynolds (1976), recorded here under the former; in Sikkink (1954) and Sikkink (1956); and in Whitehead (1969) and Whitehead (1971).

Analysis
The unit of analysis was the message pair (that is, the pair composed of an explicit message and its inexplicit counterpart). When the same messages were used in more than one investigation, results were combined. Such combined results were computed in the following cases: results recorded under Cathcart (1953, 1955) reflect results from Cathcart (1953, 1955) and from Bostrom and Tucker (1969); results recorded under “McCroskey capital punishment” reflect results from studies 1, 3, and 4 in McCroskey (1967b); results recorded under “McCroskey pro-education” reflect results from studies 1, 4, and 5 in McCroskey (1967b) and McCroskey (1970).[i]  Some designs used multiple messages but did not report results separately, and so were treated as having only one message (Berger 1988, second preliminary study and main study; Whitehead 1969, 1971); the consequence is that the present analysis underrepresents any message-to-message variability in these data.
The individual correlations (effect sizes) were initially transformed to Fisher’s zs; the zs were analyzed using random-effects procedures described by Shadish and Haddock (1994), with results then transformed back to r. A random-effects analysis was employed in preference to a fixed-effects analysis because of an interest in generalizing across messages.

Meta-analysts of message effects research face a circumstance parallel to that of primary researchers whose designs contain multiple instantiations of message categories. Such multiple-message designs can be analyzed treating messages either as a fixed effect or as a random effect. The relevant general principle is that replications should be treated as random when the underlying interest is in generalization. This reflects the fact that fixed-effects and random-effects analyses test different hypotheses: a fixed-effects analysis tests a hypothesis concerning whether the responses to a fixed, concrete group of messages differ from the responses to some other fixed, concrete group of messages, whereas a random-effects analysis tests whether responses to one category of messages differ from responses to another category of messages (see, e.g., Jackson 1992: 110). A meta-analysis involves a collection of replications (parallel to the message replications in a multiple-message primary research design), and similar considerations (including whether the analyst is interested in generalization) bear on the choice between a fixed and a random-effects meta-analysis (for some discussion, see Jackson 1992: 123; Shadish & Haddock 1994). In the present review, the interest is naturally not in the concrete messages studied by past investigators, but in the larger classes of messages of which the studied messages are instantiations; hence a random-effects analysis was the appropriate choice. In a random-effects analysis, the confidence interval around an obtained mean effect size reflects not only the usual human-sampling variation, but also between-studies variance; this has the effect of widening the confidence interval over what it would have been in a fixed-effects analysis (see Shadish & Haddock 1994: 275).

3. Results
Persuasion Effects
Details for each included case appear in Table 1. Effect sizes were available for 23 cases with a total of 5,358 participants. Across all 23 cases, the mean correlation was .064 [Q(22) = 60.2, p<.001]; the 95% confidence interval for this mean was .014, .114, indicating a significant persuasive advantage for messages providing information-source citations.

There were 13 cases (N = 2,106) involving the individual manipulation of information-source citation. Across these cases, the mean correlation was .073 [Q(12) = 23.1, p<.05]; the 95% confidence interval was .018, .128
.
There were 10 cases (N = 3,252) involving the joint manipulation of information-source citation and another message feature. Across these cases, the mean correlation was .050 [Q(9) = 37.1, p<.001]; the 95% confidence interval was -.043, .144.

Credibility Effects
Details for each included case appear in Table 2. Effect sizes were available for 10 cases with a total of 2,601 participants. Across all 10 cases, the mean correlation was .077 [Q(9) = 81.0, p<.001]; the 95% confidence interval was -.053, .206.

There were 4 cases (N = 553) involving the individual manipulation of information-source citation. Across these cases, the mean correlation was .169 [Q(3) = 10.9, p<.05]; the 95% confidence interval was .028, .311.

There were 6 cases (N = 2,048) involving the joint manipulation of information-source citation and another message feature. Across these cases, the mean correlation was .009 [Q(5) = 69.1, p<.001]; the 95% confidence interval was -.170, .188.

4. Discussion
General Effects
Characterized very broadly, these results suggest that advocates have little to fear from explicitly identifying their information sources. For studies individually manipulating information-source citation, messages with more explicit argumentative support are significantly more credible and significantly more persuasive than their less explicit counterparts.

An Implicit Limiting Condition
One might plausibly suppose that the effects (on persuasiveness and credibility) of identifying one’s information sources will depend in part on the nature of those sources. Two advocates who are equally explicit about their supporting sources might find different effects if one advocate’s sources are plainly well-qualified and trustworthy where the other’s are not.
The extant research literature does not provide extensive evidence that bears on this supposition, but two points can appropriately be made. First, in the great bulk of the research reviewed here, the sources identified in the more-explicit messages were ones likely to have been perceived as relatively high in credibility. In some cases, investigators pretested possible sources before constructing their experimental materials; for example, Bettinghaus (1953) used information sources identified in pretesting as persons thought competent to render judgments in the topic area. Investigators have commonly not intentionally sought to invoke palpably weak information sources. Thus there may implicitly be a limiting condition on the observed general effects, specifically, that persuasion-and credibility-enhancing effects of explicit source identification obtain only when the identified sources are of sufficiently high quality.
Second, the few studies that have varied the apparent quality of the identified sources have not produced consistent effects. Luchok and McCroskey’s (1978) results suggested that citing poor-quality information sources would inhibit persuasion (compared to not providing source citations); however, in Cronin’s (1972) study, citing low-credibility information sources was more persuasive than not citing any information sources.[ii]
At a minimum, then, one may say that the observed positive effects on credibility and persuasiveness obtain at least when the identified sources are recognizably sound. It is not yet clear whether there are specifiable general circumstances under which such positive effects might obtain with poorer information sources. Future research might usefully be directed at clarifying this potential limiting condition

Individual and Joint Effects
The best evidence for the effect of a given message variation obtains in designs in which that variation is manipulated independently of other message variations. In this research area, however, a number of studies have jointly manipulated information- source citation and other message properties (commonly capturing such joint variation under the general heading of “evidence”). Such designs, of course, obscure the possible causal mechanisms for any observed effects. In the research reviewed here, the observed mean effects (on credibility and persuasion) of such joint-manipulation designs are not dependably different from those of individual-manipulation designs, though the joint-manipulation mean is smaller and (unlike the individual-manipulation mean) is not dependably different from zero. Thus with respect to the research question of interest here – that is, the question of the effects of variation in information-source citation – the best evidence in hand (the evidence from individual-manipulation studies) indicates that both persuasiveness and credibility are significantly enhanced by information-source citation.

But these findings also speak to the research practice of jointly manipulating several message variables in this confounded way. Such quasi-experimental designs can be attractive for various reasons. In the early stages of research, uncertainty about possible mechanisms might recommend casting one’s net widely. For field (as opposed to laboratory) experiments, quasi-experimental designs may be more practical (e.g., Gonzales, Aronson, & Costanzo 1988; Reynolds, West, & Aiken 1990). More generally, manipulating a suite of message features can appear to promise stronger effects: one might expect relatively larger impact by contrasting two messages that vary in several features (for instance, comparing a message that lacks both quantitative specificity and information- source citations against a parallel message that is both quantitatively more explicit and provides citations to the sources of its information) rather than just one feature. Interestingly enough, however, in the limited data afforded by this research area, there is no evidence of such enhanced impact. This concretely illustrates that the effects of joint manipulations are not necessarily the sum of the effects expected from the individual manipulations, and indeed may not be larger than the effect of a single manipulation. Insofar as experimental design in persuasion effects research is concerned, then, the lesson is that the manipulation of a suite of message features does not necessarily enhance effect size.

Explaining the Observed Effects
Credibility enhancement. One appealing possible explanation of the observed effects is that explicit identification of information sources enhances the communicator’s credibility, which then leads to enhanced persuasion. Such a process would presumably involve receivers’ invoking a credibility heuristic, in which the apparent credibility of the communicator is used as a basis for assessing the advocated view (see, e.g., Chaiken 1987; Petty & Cacioppo 1986).
This explanation leads to the expectation that communicators initially low in credibility might enjoy greater impact from explicit source identification than would high-credibility communicators. High-credibility communicators might not enjoy so much credibility enhancement from explicitly identifying their sources as would low-credibility communicators (because of ceiling effects), and so they might not obtain so much greater persuasive impact.
Evidence relevant to this expectation can potentially be obtained from research designs varying both initial communicator credibility and information-source identification. A number of studies have used designs of this sort, though commonly these do not provide sufficient quantitative information to permit useful meta-analytic treatment; however, it is possible to consider simply the direction of effect observed in such studies. As a broad overview, it appears that there is not a striking difference between highand low-credibility communicators in the character of the observed effects of information-source-citation variations on either persuasive outcomes or perceived credibility.

Table 1 – Persuasion Effects

Table 2 – Credibility Effects

With respect to persuasive effects, for communicators initially high in credibility, a number of studies have indicated that messages citing information sources have some persuasive advantage over those without such citations (Harte 1972, Experiment 1; McCroskey capital punishment; McCroskey pro-education; McCroskey revised capital punishment; McCroskey revised education), but several studies have reported effects in directions favoring messages without citations (Harte 1972, Experiment 2; Hayes 1966; Luchok & McCroskey 1978; McCroskey con-education). Similarly, for communicators initially low in credibility, in several cases messages with explicit citations have been more persuasive than their nonexplicit counterparts (Luchok & McCroskey 1978; McCroskey capital punishment; McCroskey pro-education; McCroskey revised capital punishment; McCroskey revised education), but in a number of cases the opposite direction of effect has been observed (Harte 1972, Experiment 1; Harte 1972, Experiment 2; Hayes 1966; McCroskey con-education). That is, the pattern of effects does not display the expected greater superiority of information-source citation for low-credibility communicators.
Concerning credibility perceptions, for communicators initially high in credibility, a number of studies have indicated that messages with information-source citations lead to more positive credibility judgments than do messages without such citations (Fleshler, Ilardo, & Demoretcky 1974; McCroskey pro-education; McCroskey con-education; McCroskey revised education), but several other studies have reported mixed effects or effects favoring messages without explicit citations (Harte 1972, Experiment 1; Harte, 1972 Experiment 2; Hayes 1966; McCroskey capital punishment). Similarly, for communicators initially low in credibility, some studies report that messages with information-source citations enhance perceived credibility more than do messages without such citations (Fleshler, Ilardo, & Demoretcky 1974; Hayes 1966; McCroskey pro-education; McCroskey con-education; McCroskey revised education), but other cases favor messages without such citations or report mixed directions of effect (Harte 1972, Experiment 1; Harte 1972, Experiment 2; McCroskey capital punishment). Again, the pattern of effects does not suggest that low-credibility communicators enjoy some marked advantage over high-credibility communicators in the impact of information-source citations on credibility perceptions.
Thus variations in information-source citation do not seem to have dramatically different effects on the perceived credibility of, or the persuasiveness of, communicators initially high in credibility and those initially low. This research evidence is limited in a number of important ways (there are few relevant cases, effect sizes are not available, and nearly all the studies involve confounded designs), so one ought not make too much of what is in hand; future research could plainly be useful in clarifying the relevant relationships. But at a minimum the evidence to date does not give substantial encouragement to the supposition that the effects of information-source-citation variations depend in some crucial way on the communicator’s initial level of credibility. This, in turn, suggests that credibility enhancement may not be the causal mechanism by which information-source citation enhances persuasion.

Argument enhancement. An alternative possible account is that information-source citation directly enhances belief in the relevant supporting argument, thereby making the message more persuasive. That is, quite apart from any effects that such citation might have on perceptions of the communicator’s credibility, explicit source identification could enhance the persuasiveness of the supporting argumentation. For instance, a receiver might reason that a particular supporting argument is more likely to merit belief given the identification of the source of some information invoked by the argument. Thus the impact of the supporting argument might itself directly be enhanced by such explicitness, without any intervening step involving enhanced perceptions of the communicator’s credibility. From this vantage point, the observed credibility-enhancement effect of information-source citation is epiphenomenal, that is, not implicated in bringing about the observed effects on persuasiveness.
This explanation underscores the importance of research focussed on identification of specific argument features that enhance the impact of individual arguments (and thus the impact of the messages in which they appear). One well-known body of research that might appear to bear on this question is elaboration likelihood model research concerning the role that variation in “argument strength” plays in persuasion (e.g., Petty, Cacioppo, & Goldman 1981). But as several commentators have noted (e.g., Areni & Lutz 1988), this research has not specified the properties that make specific arguments relatively more or less persuasive. The present results suggest that information-source citation might be a candidate worthy of closer examination.
But there are at least two different means by which information-source citation could directly bolster the persuasiveness of supporting arguments. One possibility is that the effect arises through the receiver’s careful scrutiny of the source-identification material; if this is the underlying process, then identification of low-credibility sources might diminish persuasiveness (because the receiver’s close examination of the explicit identification material will reveal the source’s weaknesses). A second possibility is a more heuristic-like process, in which the mention of an information source is taken as a sign of the merit of the argument, in a way that does not necessarily involve careful attention to the argumentative details; if this is the underlying process, then even identification of low-credibility information sources might enhance persuasiveness (that is, citing any information source may be taken as an indication of the argument’s being worthy of belief).[iii]

5. Conclusion
Messages with more explicit identification of their information sources are significantly more credible and significantly more persuasive than their less explicit counterparts. Additional research will be needed to identify the limits of the observed effects (circumstances under which the effects do not occur, or are reversed) and to explain how and why the effects arise. But as a rule, advocates can appropriately be advised, on both normative and instrumental grounds, to explicitly articulate their argumentative support in this way.

NOTES
[i] The results recorded under McCroskey con-education are from McCroskey (1970); the results recorded under McCroskey revised capital punishment are from McCroskey (1967b, Study 5).
[ii] Warren’s (1969) design varied the credibility of information sources, and Dresser’s (1962, 1963) design varied both the credibility of information sources and the relevance of the provided material to the claims advanced; neither, however, contained a no-source-citation condition, and thus these studies could not provide evidence about the relative persuasiveness of leaving information sources uncited versus citing low-credibility sources.
[iii] Thanks to Sally Jackson for suggesting this possibility

REFERENCES
Anderson, D. (1958). The effect of various uses of authoritative testimony in persuasive speaking (Master’s thesis, Ohio State University, 1958).
Areni, C. S., & Lutz, R. J. (1988). The role of argument quality in the elaboration likelihood model. In: M. J. Houston (Ed.), Ad-vances in Consumer Research (Vol. 15, pp. 197-203). Provo, UT: Association for Consumer Research.
Babich, R. M. (1971). Perceived informativeness related to attitude change in topic, evidence, and source variant messages (Doctoral dissertation, University of Colorado, 1971). Dissertation Abstracts International 32 (1972), 4147A. (University Microfilms No. AAG-723621).
Berger, K. A. (1988). The effect of report, inference, and judgment messages in the absence and presence of a source on attitudes and beliefs (Doctoral dissertation, New York University, 1988). Dissertation Abstracts International 49 (1988), 1205A. (University Microfilms No. AAG-8814669).
Bettinghaus, E. P., Jr. (1953). The relative effect of the use of testimony in a persuasive speech upon the attitudes of listeners (Master’s thesis, Bradley University, 1953).
Bostrom, R. N., & Tucker, R. K. (1969). Evidence, personality, and attitude change. Speech Monographs 35, 22-27.
Cathcart, R. S. (1953). An experimental study of the relative effectiveness of selected means of handling evidence in speeches of advocacy (Doctoral dissertation, Northwestern University, 1953). Dissertation Abstracts 14 (1954), 416. (University Microfilms No. AAG-0007021).
Cathcart, R. S. (1955). An experimental study of the relative effectiveness of four methods of presenting evidence. Speech Monographs 22, 227-233.
Chaiken, S. (1987). The heuristic model of persuasion. In: M. P. Zanna, J. M. Olson, & C. P. Herman (Eds.), Social Influence: The Ontario Symposium, Vol. 5 (pp. 3-39). Hillsdale, NJ: Lawrence Erlbaum.
Cowdin, H. P. (1968). Development of persuasive messages based on the belief patterns of the audience, and the effects of authorities and dogmatism on change of beliefs (Doctoral dissertation, University of Iowa, 1968). Dissertation Abstracts 29 (1968), 1858A. (University Microfilms No. AAG-6816793).
Cronin, M. W. (1972). An experimental study of the effects of authoritative testimony on small-group problem-solving discussions (Doctoral dissertation, Wayne State University, 1972). Dissertation Abstracts International 33 (1973), 6485A. (University Microfilms No. AAG-7312497).
Dresser, W. R. (1963). Effects of “satisfactory” and “unsatisfactory” evidence in a speech of advocacy. Speech Monographs 30, 302-306.
Van Eemeren, F. H., Grootendorst, R., Jackson, S., & Jacobs, S. (1993). Reconstructing Argumentative Discourse. Tuscaloosa, AL: University of Alabama Press.
Fleshler, H., Ilardo, J., & Demoretcky, J. (1974). The influence of field dependence, speaker credibility set, and message documentation on evaluations of speaker and message credibility. Southern Speech Communication Journal 39, 389-402.
Gilkinson, H., Paulson, S. F., & Sikkink, D. E. (1954). Effects of order and authority in an argumentative speech. Quarterly Journal of Speech 40, 183-192.
Gonzales, M. H., Aronson, E., & Costanzo, M. A. (1988). Using social cognition and persuasion to promote energy conservation: A quasi-experiment. Journal of Applied Social Psychology 18, 1049-1066.
Harte, T. B. (1972). The effects of initial attitude and evidence in persuasive communication (Doctoral dissertation, University of Illinois at Urbana-Champaign, 1972). Dissertation Abstracts International 34 (1973), 890A. (University Microfilms No. AAG-7317234).
Harte, T. B. (1976). The effects of evidence in persuasive communication. Central States Speech Journal 27, 42-46.
Hayes, H. B. (1966). Source credibility and documentation as factors of persuasion in international propaganda (Doctoral dissertation, University of Iowa, 1966). Dissertation Abstracts 27 (1966), 450A. (University Microfilms No. AAG-6607209).
Hayes, H. B. (1971). International persuasion variables are tested across three cultures. Journalism Quarterly 48, 714-723.
Holtzman, P. D. (1966). Confirmation of ethos as a confounding element in communication research. Speech Monographs 33, 464-466.
Hovland, C. I., & Mandell, W. (1952). An experimental comparison of conclusion-drawing by the communicator and by the audience. Journal of Abnormal and Social Psychology 47, 581-588.
Jackson, S. (1992). Message Effects Research: Principles of Design and Analysis. New York: Guilford Press.
Jackson, S., & Jacobs, S. (1980). Structure of conversational argument: Pragmatic bases for the enthymeme. Quarterly Journal of Speech 66, 251-265.
Johnson, B. T. (1989). DSTAT: Software for the Meta-Analytic Review of Research Literatures. Hillsdale, NJ: Lawrence Erlbaum.
Kardes, F. R. (1988). Spontaneous inference processes in advertising: The effects of conclusion omission and involvement on persuasion. Journal of Consumer Research 15, 225-233.
Kellermann, K. (1980). The concept of evidence: A critical review. Journal of the American Forensic Association 16, 159-172.
Kilcrease, P. P. (1977). An experimental investigation of Fishbein and Ajzen’s model of behavioral prediction, their principles of change, and the effect of the use of evidence in a persuasive communication (Doctoral dissertation, Louisiana State University, 1977). Dissertation Abstracts International, 38 (1978), 3803A-3804A. (University Microfilms No. AAG-7728684).
Luchok, J. A. (1973). The effect of evidence quality on attitude change, credibility, and homophily. Unpublished master’s thesis, West Virginia University.
Luchok, J. A., & McCroskey, J. C. (1978). The effect of quality of evidence on attitude change and source credibility. Southern Speech Communication Journal 43, 371-383.
McCroskey, J. C. (1966). Experimental studies of the effects of ethos and evidence in persuasive communication (Doctoral dissertation, Pennsylvania State University, 1966). Dissertation Abstracts 27 (1967), 3630A. (University Microfilms No. AAG-6705945).
McCroskey, J. C. (1967a). The effects of evidence in persuasive communication. Western Speech 31, 189-199.
McCroskey, J. C. (1967b). Studies of the effects of evidence in persuasive communication. East Lansing, MI: Michigan State University Speech Communication Research Laboratory, report SCRL 4-67.
McCroskey, J. C. (1969). A summary of experimental research on the effects of evidence in persuasive communication. Quarterly Journal of Speech 55, 169-176.
McCroskey, J. C. (1970). The effects of evidence as an inhibitor of counter-persuasion. Speech Monographs 37, 188-194.
McCroskey, J. C., & Dunham, R. E. (1966). Ethos: A confounding element in communication research. Speech Monographs 33, 456-463.
O’Keefe, D. J. (1997). Standpoint explicitness and persuasive effect: A meta-analytic review of the effects of varying conclusion articulation in persuasive messages. Argumentation and Advocacy 34, 1-12.
Ostermeier, T. H. (1966). An experimental study on the type and frequency of reference as used by an unfamiliar source in a message and its effect upon perceived credibility and attitude change (Doctoral dissertation, Michigan State University, 1966). Dissertation Abstracts 27 (1967), 2216A. (University Microfilms No. AAG-6614159).
Ostermeier, T. H. (1967). Effects of type and frequency of reference upon perceived source credibility and attitude change. Speech Monographs 34, 137-144.
Petty, R. E., & Cacioppo, J. T. (1986). Communication and Persuasion: Central and Peripheral Routes to Attitude Change. New York: Springer-Verlag.
Petty, R. E., Cacioppo, J. T., & Goldman, R. (1981). Personal involvement as a determinant of argument-based persuasion. Journal of Personality and Social Psychology 41, 847-855.
Reinard, J. C. (1984). The role of Toulmin’s categories of message development in persuasive communication: Two experimental studies on attitude change. Journal of the American Forensic Association 20, 206-223.
Reinard, J. C. (1988). The empirical study of the persuasive effects of evidence: The status after fifty years of research. Human Communication Research 15, 3-59.
Reinard, J. C. (1994). The persuasive effects of assertive evidence. In: M. Allen & R. W. Preiss (Eds.), Prospects and Precautions in the Use of Meta-Analysis (pp. 127-157). Dubuque, IA: Brown and Benchmark.
Reinard, J. C., & Reynolds, R. A. (1976, November). An experimental study of the effects of Toulmin’s pattern for argument development on attitude change. Paper presented at the meeting of the Western Speech Communication Association, San Francisco, CA. (ERIC document number ED133782).
Reynolds, K. D., West, S. G., & Aiken, L. S. (1990). Increasing the use of mammography: A pilot program. Health Education Quarterly 17, 429-441.
Rosenthal, R. (1991). Meta-Analytic Procedures for Social Research (rev. ed.). Newbury Park, CA: Sage.
Shadish, W. R., & Haddock, C. K. (1994). Combining estimates of effect size. In: H. Cooper & L. V. Hedges (Eds.), Handbook of Research Synthesis (pp. 261-281). New York: Russell Sage Foundation.
Sikkink, D. E. (1954). An experimental study of the effects on the listener of anti-climax order and authority in an argumentative speech (Doctoral dissertation, University of Minnesota, 1954). Dissertation Abstracts 14 (1954), 2157. (University Microfilms No. AAG-0009622).
Sikkink, D. E. (1956). An experimental study of the effects on the listener of anticlimax order and authority in an argumentative speech. Southern Speech Journal 22, 73-78.
Wagner, G. A. (1958). An experimental study of the relative effectiveness of varying amounts of evidence in a persuasive communication (Master’s thesis, Mississippi Southern College, 1958).
Warren, I. D. (1969). The effect of credibility in sources of testimony on audience attitudes toward speaker and message. Speech Monographs 36, 456-458.
Whitehead, J. L., Jr. (1969). An experimental study of the effects of authority-based assertion (Doctoral dissertation, Indiana University, 1969). Dissertation Abstracts International 30 (1970), 5101A-5102A. (University Microfilms No. AAG-7007518).
Whitehead, J. L., Jr. (1971). Effects of authority-based assertion on attitude and credibility. Speech Monographs 38, 311-315.