Computational Uses of Philosophical 
Dialogue Theories 

DAVIDMoORE& 

DAVEHoBBES 

Abstract: The research discussed in this pa-
per concerns an investigation of logical dia-
logue games as a vehicle for enhanced hu-
man-computer communication. The need for 
a dialogue capability in computer systems is 
argued, and a prima facie case made for dia-
logue games, in particular Mackenzie's "DC", 
as a suitable dialogue model. Empirical work 
concerning DC is outlined, and strategies for 
adoption by the computer are presented. A 
user interface, software architecture and con-
ceptualised example are discussed. Thc hopc 
is that the paper will help bring together two 
groups of workers-philosophers of dialogue 
and designers of computer systems-to the 
mutual benefit of each. 

Leeds Metropolitan 
University 

Resume: La recherche discutee dans 
eet article porte sur I' investigation des jeux 
de dialogue logique employes pour faciliter 
la communication humaine avec les 
ordinateurs. On soutient que I'aptitude au 
dialogue des systemes informatiques est6 
un reel besoin etqu'a premiere vue lesjeux 
de dialogue y repondent. En particulier, Ie 
"DC" de Mackinzief semble offrir un 
modele approprie de dialogue. On trace 
les grandes lignes des recherches 
empiriques sur Ie DC et I'on propose des 
stategies pour I' appliquer aux ordinateurs. 
On discute de I' interface pour l'usager, de 
laconfiguration du logiciel et d'un exemple 
conceptualise. On espere que cet article 
encouragera un rapprochement 
mutuellement avantageux entre deux 
groupes de travailleurs: les philosophes du 
dialogue et les concepteurs des systemes 
informatiques. 

Keywords: logical dialogue games, DC, Intelligent tutoring systems 

1. Introduction 

The aim ofthis paper is to alert people working in what can be called the philosophy 
of dialogue (Walton 1989) to computational utilisations of theoretical insights gained 
in this field, and hence perhaps to pave the way towards a fruitful collaboration be-
tween the fields. The paper will therefore start by arguing the need for a dialogue 
capability within interactive computer systems. It will then outline some of the diffi-

© Informal Logic Vol. 18, Nos. 2&3 (1996): pp. 131-163. 



132 David Moore & Dave Hobbes 

culties engendered by attempting to provide dialogue, and briefly review the theoreti-
cal foundations of some ofthe attempts to address these difficulties. Since the par-
ticular interest of the current authors is in the utilisation oflogical dialogue games, in 
particular the system "DC" (Mackenzie I 979a), our work in these areas will be pre-
sented in rather more detail in the remainder of the paper. 

2. The need for dialogue within interactive computer systems 

The ubiquity of the computer, and the corresponding advent of end-user personal com-
puting, has made it vital that end-users are able successfully to operate the systems 
with which they are working; failure or refusal to do so can mean a lack of satisfac-
tion for the user, and of success for the system. This is the problem which the field of 
Human Computer Interaction (HCI) seeks to address (e.g., Shneidennan 1992). One 
area which receives much attention is that of dialogue design. Despite some tennino-
logical confusion in the area (Booth 1989), it is safe to say that dialogue design en-
compasses two broad issues. The first concerns the manner in which the user will 
interact with the machine, whether, for example, they will be presented with a menu of 
options, a command line interface, or some fonn of question-answer interaction. The 
second area concerns the attempt to emulate dialogue of the sort engaged in by human 
conversants, in which topics of mutual concern are discussed, and in which each con-
versant has substantive points to make; the interest is in the content and fonn ofthe 
dialogue, rather than the manner ofits physical implementation. It is the second area 
with which we are chiefly interested. 

One type ofinteractive computer system which would benefit greatly from such 
dialogue is a computer based learning (CBL) system. Whilst it is not without its crit-
ics (e.g., Ridgway 1988), the use ofCBL in education offers advantages in tenns of 
progressing at the pace of the individual student and of allowing the student to experi-
ment and interact without the fear of ridicule from his tutor or peers (Mushrush 1990), 
and of enabling learning to be customized to the needs of the individual student (Elsom-
Cook 1988). As O'Shea and Self (1983) put it, the aim is for each student to have their 
own Aristotelian tutor. A dialogue capability should be seen, it can be argued, as a 
fundamental part of such systems (cf. Baker 1994). For the importance of discussion 
and debate in education is frequently stressed at primary (e.g., National Curriculum 
Council 1990a), secondary (e.g., National Curriculum Council 1990b ), and tertiary 
(e.g., Garrison 1991) levels. And this, together with the practical difficulties of fa-
cilitating one-to-one discussion in education (e.g., Galton and Simon 1980), suggest 
great potential for CBL systems to make a major practical contribution in this regard. 
This in tum suggests a need for interactive dialogue in such systems. 

Another type of system that would benefit from such dialogue is the so-called 
"expert system" (e.g., see Jackson 1990). These systems seek to encapsulate expert 
knowledge about some typically constrained domain, and to engage with users in "con-
sultation dialogues" (Bench-Capon 1990) with a view to solving specific problems 



Computational Uses of Philosophical Dialogue Theories 133 

within that domain. An explanation facility is regarded as an important component of 
expert systems, but current provision has been criticised as being hard for users to 
understand, verbose, and not matched to the users' expectations or prior knowledge, 
and "interactive argument" has been proposed as a more useful model of interaction 
(Bench-Capon et al. 1991). Another type of system which would benefit from dia-
logue capability is the on-line help system. Empirical work connected with the 
EUROHELP project, for example, showed that more than two-thirds of answers to 
help queries were followed up with further queries, conjectures, or counter-asser-
tions, clearly suggesting the need for dialogue (Pilkington 1992a). 

3. The Dialogue Management Problem 

It has turned out to be very difficult to emulate dialogue computationally, and there is 
a great need for further research: "There remains a substantial amount of foundation 
work to be done in order to make extended models of conversation that are both 
computationally tractable and formally sound" (Finkelstein and Fuks 1990). A natural 
point of departure for such foundation work is research involving human dialogue 
(Girle 1986). The issue of how humans conduct dialogue is, however, itself highly 
controversial. Shiffrin (1985), for example, claims that discourse analysis is "so in-
ternally differentiated that its practitioners [fail] to recognise any common core". 

This controversy is, as one might expect, reflected in current attempts to build 
computational dialogue systems (cf. Stenton 1988). Reichman (1985), for example, 
has devised a system in which discourse is characterised as a hierarchical organisa-
tion of related types of utterances. A "context space" is delineated together with a 
specific set of conversational moves in that space, such as "support" or "conclusion". 
A set of discourse rules have been developed, which seek to transform Gricean max-
ims into a set of operational rules governing the interpretation and generation of 
"maxim-abiding" conversation. 

Reichman is explicitly opposed to basing computational systems on conversa-
tional analysis. Frohlich and Luff (1990), on the other hand, do seek to ground a dia-
logue system on ethno-methodological concepts, and claim that such a model "will 
prove more appropriate than any other for the design of human-computer conversa-
tions". Adopting what might be seen as an intermediate position, Cawsey (1990) uti-
lises ideas from both conversation and discourse analysis, and Finkelstein and Fuks 
(1990) seek to integrate work from what are often seen as the contending approaches 
of conversation analysis and discourse analysis. 

Levin and Moore (1977, cf. Mann 1988) claim to have empirically identified sev-
eral dialogue games, conceived as identifiable types of "systematic interaction" in 
which conversants engage. These games are used as the basis for a computational 
"process model" ofthe comprehension of dialogue utterances, and are also used by 
Elsom-Cook (1985) and Baker (\989) in their intelligent tutoring systems in the 
realms of, respectively, the programming language LISP, and music. 



134 David Moore & Dave Hobbes 

Carlson (1983, 1984a, 1984b) develops a complex theory of idealized conversa-
tion. Carlson speculates as to the utility ofthe theory in computational contexts, and 
the model has in fact been used as the basis for the implementation of the intonation 
interpretation module of a speech-understanding system (Hoepelman et al 1991), 
though it is admitted that "the dialogue model and the dialogue rules are still very 
preliminary". 

There is, then, a plethora of approaches to computational dialogue, each grounded 
to some extent on an underlying theory of human discourse. In our current research 
we have adopted the policy, rather than await the resolution of the controversy, of 
selecting a discourse model and putting its potential to the test. If the dialogues gen-
erated as a result of its use are deemed to represent satisfactory dialogue, and thus to 
offer an improvement over current facilities, the model will be deemed to that extent 
successful. If not, the model will be deemed to that extent unsuccessful. The model to 
be adopted is derived from work on "logical dialogue games" (e.g. Walton 1984). The 
central claim to be investigated in our research, therefore, is that such games can 
provide a suitable framework for a computer-based dialogue system. For this claim to 
be of interest and hence worthy of investigation, there should be some prima facie 
case for the adoption of such games. This case will be made in the next section. 

4. Logical dialogue games 

What will here be referred to as "logical dialogue games" in fact have a range of 
different nomenclatures in the literature, such as "formal games of dialogue", "logi-
cal structures", "dialectical games" (Walton 1984), and "dialectical systems" (e.g. 
Hamblin 1987, Mackenzie 1979b). The games are also defined in somewhat different 
ways. Hamblin (1971) defines a dialectical system as a triple <P,L,K>, where P is a 
set of participants, L a set of locutions, and K a set oflegaJ dialogues defined within a 
dialogue by a set of rules. Mackenzie (1979b) defines a "dialectical system" as a 
triple <P,L,R>, where P is a set of participants, L a set of locutions, and R a set of 
rules which define a set K of legal dialogues. Walton (1984) suggests that within the 
framework of a logical dialogue game, an argument can be seen as a set of locutions, 
indexed to a participant in the game, where the participant may advance a locution 
only according to certain dialogue rules; he then defines a dialectical system as such 
a game, with the addition of a "strategic component defining win-loss requirements" 
(Walton 1984). Elsewhere (1989) he suggests, in a definition reminiscent of 
Hamblin's, that" a game of dialogue is a triple: (1) a set of players, (2) a set of moves, 
(3) a set of rules". The names and definitions are somewhat different from each other, 
therefore, but the general notion is clear: a logical dialogue game is a set of rules, 
regulating the participants as they make moves in the dialogues. These rules will leg-
islate as to permissible sequences of moves, and also as to the effect of moves on 
participants' "commitment stores", conceived as records of statements made or ac-
cepted. 



Computational Uses 0/ Philosophical Dialogue Theories 135 

The motivation behind the development of such games is that the study of argu-
mentation in such contexts will supplement apparent defects in the application of 
fonnallogic (Walton 1985, Barth and Martens 1982). In particular, the concern is in 
many cases with modelling (and hence helping to avoid) fallacious argumentation. 
Thus Girle (1986): "When a dialogue becomes illegal, so the theory goes, fallacies 
occur". An implication of this is that, despite some apparent ambivalence with regard 
to the status of the theory (cf. Moore 1993), it can be seen as a prescription for 
reasonable, rather than descriptive of actual, dialogue: " ... at one level ... the prag-
matics of dialogue involves the prescriptive study of what is fair and reasonable in 
argument and criticism" (Walton 1985, cf. Finkelstein and Fuks 1990). 

Logical dialogue games, as so characterised, have a number of apparent attrac-
tions from the point of view oftheir utilisation within a computer dialogue system. 
First, given that the games purport to be models of "what is fair and reasonable in 
argument and criticism" (Walton 1985), then constraining both computer and user to 
such a game will, if the game is valid, yield "fair and reasonable" dialogue, and thus 
satisfy the desiderata discussed in section 2. A second part of the case for the adop-
tion of logical dialogue games relates to claims made for their educational advan-
tages. Walton, for example, suggests that "the circular staircase of dialectic can be an 
assent out of the cave offallacy, from the darkness towards the light" (Walton 1984), 
and Hartley and Hintze (1990) suggest as educational advantages of a logical dia-
logue game approach within CBL that the student can probe and question the tutor, and 
express his own viewpoint. 

A further argument is that the discipline imposed on players by the games is likely 
to make more tractable the task of providing for dialogue via machine. Several factors 
contribute to this likelihood. The first concerns the notion of dialogue rules. The aim 
would be to utilise these generic rules to establish the legality of student input, and to 
assist infonnulating a response by restricting attention to the set oflegal moves, and 
thus to guide the dialogue beyond a single question/answer interaction. The computa-
tional attractiveness of the dialogue game rules is enhanced by a consideration of 
their nature. For the rules of logical dialogue games are characteristically tight; Mac-
kenzie's (1 979a) rules, for example, prescribe no more than three types of responses 
to an incoming move type, thus promising to ease the task of checking for legality, 
and to reduce both the search problem for the computer's rejoinder and the strategic 
issue of selection between competing alternatives. 

The games' concern with modelling commitment rather than belie/is also attrac-
tive from a computational perspective. Commitments are incurred during the dia-
logue, in line with clearly stated commitment rules, and the resulting "commitment 
store" is seen not as a psychological model of memory, but rather as akin to a publicly 
inspectable set of statements recorded on a slate (Mackenzie 1979a, Hamblin 1970). 
This separation of commitment from belief has the advantage that one can allow for 
machines to argue without having to concede that they have beliefs (Mackenzie 1979a). 
Equally important, the explicit nature of the commitment function intuitively sug-
gests computational feasibi lity. 



136 David Moore & Dave Hobbes 

The characterisation of dialogue moves also appears advantageous from the com-
putational point of view, for several reasons. First, the content of a move is generally 
restricted to one "locution", i.e. a statement together with a statement operator (e.g., 
assert, question, withdraw). Although this length restriction has the consequence that 
certain dialogues, e.g., the cannibalism dialogue of Woods and Walton (1982a), cited 
approvingly by Walton himself (1989), would fall outside the scope of such systems, 
it has the computational advantage of avoiding complexities such as deciding on a 
practical length for turns (Clark and Schaefer 1989), or formulating a tum-length 
control policy (Frohlich and Luff 1990), and difficulties involving moves spanning 
more than one speaker tum (Reichman 1985). 

Further, the restricted range of move types allowed by the games potentially makes 
it possible to use a menu scheme for user input. In addition, the rule bound nature of 
the moves enables relatively straightforward checks on the legitimacy of input. The 
strictures of the rules can be expressed as move type pre- and post-conditions, and 
this suggests the possibility of autonomous move type objects which have an aware-
ness of their own availability at certain stages of the game, and also of internal con-
sistency checks between the currently applicable rule and the currently satisfied pre-
conditions. 

Perhaps the main computational attraction, however, is the clarity and expressive-
ness provided by the moves. Their pre- and post-conditions relate purely to explicit 
moves and to participants' (inspectable) commitment stores, so that complications 
concomitant upon implicit moves never arise. Neither is there any requirement to 
attempt the difficult task of divining the intentionality of the maker of the move (cf. 
Reichman 1985, Mackenzie 1990), given the move's clearly defined function within 
the game. Intentionality is, in effect, embedded within the move. The difficulty of 
interpreting pragmatic content is, therefore, largely overcome: "the idea would be to 
limit formally the permissible forms of expressions so that intention in utterance 
would be unambiguous and the need for complex pragmatic parsing would be bypassed" 
(Pilkington 1992a). 

Thus moves and commitments are clear, and dependent only upon the locution 
used (Mackenzie 1979a) so that issues of iIIocutionary force, for example (Stenton 
1988), are by-passed, as are "dynamic social-psychological factors" such as polite-
ness (Reichman 1985). The possible loss of richness consequent upon this can be 
seen simply as a reflection of the different paradigm that computer-based communi-
cation offers, and greatly improves the prospects for successful computer implemen-
tation (cf. Pilkington 1992a). 

Further support for the approach's computational tractability can be found in the 
literature. It is seen by Girle (1986, cf. Stewart-Zerba and Girle 1993) and Hartley 
and Hintze (1990) as a potential means of widening the communication channel of 
CBL systems. Bench-Capon et al. (1991) see dialectic as a useful model of interac-
tion with a knowledge-based system, and utilise a dialogue manager incorporating a 
Mackenzie-style dialogue game. Finkelstein and Fuks (1990) also favour the adop-



Computational Uses of Philosophical Dialogue Theories 137 

tion of a Mackenzie-style dialogue game, enriched with certain concepts drawn from 
ethnomethodology, as the basis for a system for providing automated support for de-
veloping software specifications. Fuks et al (1989) seek to use a Hamblin-style ap-
proach to dialogue to provide automated support for legal reasoning. and Jeremaes 
(1987) to use dialogue games to characterise a logic of interaction. Dialogue game 
theory (DGT) is seen as a potentially useful addendum to the intelligent help system 
"EUROHELP" (Hartley et aI1990), and Bailey (1989) suggests the possible use of 
DGT to extend user interactions with her explanation-driven medical expert system. 
More generally, Gordon (1994) uses a Lorenzen-style dialogue game to implement a 
computational model of civil pleading, and speaks of "Computational Dialectics" as a 
new subfield of Artificial Intelligence. 

There are, then, a number of convincing reasons for the adoption oflogical dia-
logue games as the basis for the dialogue model in a computer dialogue system. There 
are also a number offactors that have been explored with regard to the design of such 
games, concerning the appropriate rule set, the nature of commitments, and whether 
the games should be competitive (see Moore (1993) for a discussion ofthese design 
parameters). Given these various design parameters, it is not surprising, perhaps, that 
a large number of alternative dialogue game systems have been proposed. Walton 
(1984), for example, discusses some 15 different games, and Barth and Krabbe (1982) 
describe a dozen or so others. The issue, as far as a computer-based dialogue system 
is concerned, therefore, would be one of deciding which of the various game types 
would be the most appropriate as a vehicle for the system. The game type currently 
forming the focus of our research will be discussed in the next section. 

5. The game "DC" 

For the purposes of the current research, a (somewhat modified version of) Macken-
zie's dialectical system "DC" (Mackenzie 1979a) has been adopted. The system al-
lows 5 move types: statement (,P', 'Q' etc., and their truth functional compounds), 
withdrawals (e.g. 'no commitment to P'), questions ('is it the case that P?'), chal-
lenges ('why is it held that P?'), and resolution demands ('resolve whether P'). There 
are 4 rules regulating commitment stores: stores are null at dialogue commence-
ment; statements by either participant are added to the stores of each; a statement P in 
response to a challenge of Q results in both P and P-> Q being added to each store; 
a challenge of P results in P being added to the store of the hearer, and P being re-
moved from, and why-P being added to, the store of the maker ofthe move. 

There are 6 dialogue rules in our amended version of the system (see Moore (1993) 
for a discussion ofthe amendments): participants may utter individual permitted lo-
cutions in tum (I\ORM); mutual commitments may not be uttered (~PSTAT); the ques-
tion P? must be answered by 'P', 'not P', or 'no commitment P' (RQUEST); 'why P?' 
must be responded to by a withdrawal of'P', a statement not under challenge by its 
speaker, or a resolution demand of any commitments of the hearer which immedi-
ately imply 'P' (R

CHAL
); resolution demands may be made only ifthe hearer is com-



l38 David Moore & Dave Hobbes 

mitted to an immediately inconsistent conjunction of statements, or withdraws or 
challenges an immediate consequent of his commitments (~SOL\'E); a resolution de-
mand must be followed by withdrawal of one of the offending conjuncts, or affirma-
tion of the disputed consequent (~SOLUTION)' 

The notions of immediate consequence and immediate inconsistency are defined 
by Mackenzie with the help of a list (V) of "preferred valid argument schemata" 
(1979b), i.e., "argument patterns immediately recognised as valid by the participants" 
(Mackenzie 1990). Immediate consequence consists of aninstantiation of one of the 
schemata. Immediate inconsistency consists either of a statement together with its 
denial, or of a set of statements Z, together with the denial of an immediate conse-
quence of Z. The contents of V are, according to Mackenzie (1979b), to be discov-
ered by empirical observation oflanguage--conditionals which participants in dia-
logue regard it as "bizarre" to deny or doubt are examples ofimmediate consequence 
conditionals, and V is built up from a study of participants' "bizarreness reactions" 
(Mackenzie 1990). 

This, then, is, in outline, the (amended) system "DC". The reasonableness of the 
adoption ofthis system can be argued for partly on negative grounds. For our initial 
aim is to utilise DC as the dialogue model in a computer system providing for educa-
tional debate, in which the computer will take up a position on a controversial issue 
diametrically opposed to that of the student, and engage the student in debate on that 
issue; this, it is held, may foster the student's debating skills and level of critical 
awareness, and make him more aware ofthe substantive issues involved (Moore 1993). 
And other documented systems within the logical dialogue games field appear not to 
be entirely appropriate for this role (Moore 1993). In particular, the Lorenz approach 
(Lorenz 1982, 1987) is argued to be asymmetric and overly deterministic, the Barth 
and Krabbe (1982) approach to be hard for the student to master and not directly 
applicable to debate, and the Hintikka approach (Hintikka and Hintikka 1982) to be 
biased towards the questioner. There are, it is argued, conceptual difficulties with the 
dark-side commitment set on which many ofW alton's systems are based (e.g., Walton 
1989), and Mackenzie's work is widely seen to represent a step forward from the 
Hamblin games upon which it is based. This is not to deny the general worth ofthese 
other systems, merely to suggest that they may not be appropriate for the specific 
requirements of a computer-based debating system. 

More positive arguments for the adoption of DC can also be cited. Walton (1984) 
argues that "the design of the rules appears practically useful", by enabling partici-
pants to enforce a "certain surface rationality in the opponent's commitment store", 
without requiring omniscience vis-a-vis the consequences of the commitments; this 
reflects Mackenzie's own view that "people are not perfect logicians" (1985). Mac-
kenzie claims to give DC a "more rigorous formulation" than given previously to 
systems used in the solution offallacies (1979a), and that his treatment of resolution 
demand goes beyond that of Hamblin (Mackenzie 1985). 



Computational Uses of Philosophical Dialogue Theories 139 

The rules ofthe system are clear, and little knowledge is needed to apply them. In 
order to assess the legality of an event within a dialogue, it is necessary to know only 
the state ofthe commitment stores, the previous event, and the "syntactic relations 
between locutions" (Mackenzie 1979a). This simplicity of the rules is also important 
from the educational point of view. For a pre-requisite of the educational use of such 
a system is that students are able to adopt its prescriptions with a minimum of cogni-
tive difficulty, and whilst the degree to which this pre-requisite is in fact met by DC is 
an empirical question, it can at least be claimed on an a priori basis that its relative 
simplicity makes the issue of DC' s suitability worthy of empirical study. 

Further, certain educational advantages can be claimed for DC, were it in fact to 
meet the pre-requisite of ease of use. The game is symmetric (Mackenzie 1985), so 
that both parties to it have the same rights and obligations, and may have to defend 
their own position. This means that the tutor can bui Id its own position (cf. Pilkington 
et al 1992), thus making the student consider views of which he was perhaps previ-
ously unaware; indeed, the two-way commitment rules of the system prevent the stu-
dent from simply ignoring the views of the tutor (and vice-versa). A further implica-
tion is that the student is allowed to challenge the tutor's position. This enables him to 
receive different explanations of that position (Finkelstein and Fuks 1990), and to 
cease to regard the tutor's views as indisputable. 

The game symmetry also implies that the student will be able to build a position 
himself. Thus student initiative can be catered to, and the student will be required to 
be consistent and to face the consequences of what he says (cf. Mackenzie 1981). He 
will also be required to support (give reasons for) views he puts forward, and in so 
doing may deepen his understanding of his adopted view by discovering hitherto "sup-
pressed" premises (Mackenzie 1979a). Commitments of both parties are made clear 
by the commitment rules, and since at any stage the number of entries in a store is at 
most twice the number of moves made to that stage in the dialogue, combinatorial 
explosions of commitments cannot occur (Mackenzie 1990). The system is non-cu-
mulative (Mackenzie I 979a), so the student is able to withdraw commitments he no 
longer wishes to adhere to, and to give non-committal responses to questions to which 
he does not know the answer. Since there is no defined end-point to the game in terms 
of, for example, maximum number of moves, the game can end in a draw, thus reflect-
ing both the fact that educationally worthwhile interaction can take place without one 
party winning the debate (Moore 1993), and Walton's view that "agnosticism should 
be permissible in reasonable dialogue" (1989). 

As well as these general arguments for the adoption of DC, it is also possible to 
point to certain precedents for its use in a computational context (see Bench-Capon 
etal (1991), Hartley and Hintze (1990), and Finkelstein and Fuks(1990». As a final 
plank in the case for adoption of DC, it is worth giving an example of the sort of 
dialogue that participants can produce by the use ofthe system. The following hypo-
thetical example from a medical context is based on a dialogue cited in Pilkington et 
al. (1992): 



140 DavidMoore & Dave Hobbes 

A: A diuretic is the appropriate treatment. 

B:Why? 
A: You have said that the symptoms include low blood pressure, high pulse, and 

low central venous pressure, and that if these systems occur together then 
heart failure is a likely candidate. 

B: You are suggesting that a diuretic is a suitable treatment for heart failure? 
A: Yes. 
B: Is the patient's fluid intake for the last six hours normal? 
A: No. 
B: Is the fluid intake low? 
A: Yes. 
B: If the fluid intake is low, then the symptoms could be caused by fluid imbal-

ance. 
A: Fluid imbalance is an alternative but less probable candidate. 
B: Is it not the case that if you were to give a diuretic and the cause of the 

symptoms is fluid imbalance, then the patient's condition would worsen? 
A: Yes. 
B:Why? 
A: A diuretic would cause further dehydration. 
B: So, you agree that a diuretic is not necessarily the best treatment? 

Other examples can be found in Finkelstein and Fuks (1989, 1990), Hartley and 
Hintze (1990), Moore (1993) and (Mackenzie 1984). Such dialogues are realistic 
and of potential educational benefit to the users. The practicality of such dialogues, 
both from the point of view of students successfully adopting the dialogue model 
(DC) and from the point of view of having a computer act as a partner in them, is the 
major issue of our current research. 

There is therefore much to be said for the proposal to adopt DC. This is not to say, 
however, that there are no weaknesses with the system. Indeed, a range of potential 
weaknesses are discussed in (Moore 1993). One problem highlighted by Mackenzie 
himself concerns the cumulative nature of challenges, for it has been argued (Woods 
and Walton 1982b) that it erroneously bans certain sequences of moves as question 
begging (cf. Mackenzie 1984). The best technical solution may be to allow, in re-
sponse to a challenge, statements which though once under challenge have been sub-
sequently uttered by the challenger, but this was rejected for the initial prototype on 
the grounds that it was likely to cause undue complexity. 

6. Assessing the utility of De within a computer dialogue system 

The proposal therefore is to adopt an amended version of DC as the framework for the 
putative computerised dialogue system. In order to assess the validity of this pro-
posal, it is necessary to establish whether DC offers a computational model of dia-
logue. It is necessary to ask therefore first whether DC is a satisfactory model of 



Computational Uses of Philosophical Dialogue Theories 141 

dialogue at all, and, second, whether the model is computationally tractable. 
The nature of the first enquiry is largely determined by the status of DC, which 

will be taken as a prescriptive model. Given this status, one thing that needs to be 
shown is that people are able to adopt and use the prescriptions to generate dialogue. 
This general requirement of prescriptive models, echoed by Walton (1989) in his talk 
of rules being clear and simple enough to be agreeable to both participants, is particu-
larly important in both education (for to be educationally viable the prescriptions 
must not be too cognitively taxing), and in a computational context (for a user must be 
aware, for example, of the significance of the menu items from which he may select). 
Whether this requirement in fact is met by DC is not something that can be settled on 
an a priori basis. It may turn out, for example, that the rules are overly complex, stand-
ing in the way of, rather than facilitating, the generation of useful argument. 

If people are able to adopt the prescriptions, it then needs to be shown that the 
resulting dialogue can be considered "fair and reasonable", in a sense to be discussed 
later. This question also needs to be settled empirically. For the issue of interest is 
not whether a philosophical expert could manipulate the rules in such a way as to 
generate dialogue which is not fair and reasonable (e.g. dialogue which is fallacious). 
The issue is, rather, whether, even if unfair dialogue can in principle be generated, it is 
in fact generated in practice. The claim, if upheld, will show only that DC tends to 
generate fair and reasonable dialogue; the stronger claim that only DC can generate 
such dialogue will not of course be demonstrated. 

Whilst the truth of the stronger claim is not needed to justify the system's compu-
tational use, some cognisance does need to be taken of the properties of the dialogue 
resulting from use of DC as compared with those of naturally occurring dialogue. 
Simply to show the dialogue is fair and reasonable is not enough, for it may differ 
from unconstrained dialogue in ways which are not acceptable. To draw an analogy 
with another prescriptive field, a medical cure will be rejected if it causes unaccept-
able side-effects, even though it may cure the original illness. 

Again, this issue cannot be settled on an a priori basis, and a comparison of DC-
generated dialogue with unconstrained dialogue will be required. This comparison 
will be important in its own right, and also in relation to the trade-off between the 
cognitive effort of adopting the DC regimen, and the gains accruing from so doing, 
for the less satisfactory the comparison is judged to be, the less advantageous the 
trade-offwill become. Carlson (1984a) suggests that computational models of dia-
logue may involve strong simplification and regimentation of ordinary human conver-
sation, and the current enquiry may be revealing vis-a.-vis the acceptability of such 
regimentation in the case of DC. It is important to remember, however, that the pur-
pose ofthe comparison is not to establish whether DC emulates ordinary dialogue, 
since it is seen as a different paradigm-cf. Barth and Krabbe's point (1982) that their 
rules are not intended to mirror just everything that happens in debate. The point is, 
rather, to seek to establish the acceptability of any alterations from ordinary conver-
sation which may be induced by adherence to the DC regimen. 



142 David Moore & Dave Hobbes 

Ifit is shown, as a result ofthe enquiries just outlined, that people are able to use 
DC to generate dialogue which is fair and reasonable and not unduly constrained in 
comparison with natural dialogue, then this would both provide supporting evidence 
of the validity ofthe model, and sanction the use of the game rules as the prescriptive 
basis for the human-computer interaction. For such interaction to be viable, it needs 
to be shown, further, that the framework offered by DC is computationally tractable. 
It needs to be shown, in particular, that a computer is able to use the rules to regulate 
debate; for the computer actually to participate in debate it would also need to be 
shown that sufficient strategic and semantic wisdom can be given to the machine, 
since DC operates at a very high level of abstraction. An affirmative finding would 
mean that the educational advantages discussed in Section 2 can potentially be achieved. 
It would also illuminate Mackenzie's claim (1981) that machines may be able to par-
ticipate in dialogue and add to Barth's claim (1982) that dialectical systems are of 
practical as well as philosophical value. 

The broad questions requiring to be investigated, therefore, are first whether the 
DC regimen is a valid prescriptive model of dialogue, and second whether the system 
is computationally practical. 

There are, however, at least two difficulties that might be raised against this pro-
posed enquiry. First, Anastasi (1982) reports the view that, given appropriate instruc-
tion, virtually anyone can achieve mastery of any instructional objective. Applying 
this to the current proposal, it might be argued that the issue of whether people can 
learn DC is vacuous, for the answer will inevitably be in the affirmative. What matters, 
though, is the ease and readiness with which the regimen can be adopted, rather than 
whether it could in principle be taught to everyone, for this will affect the relative 
advantage afforded by the system at the user interface. The question is not, therefore, 
vacuous. 

Second, it might be said that a negative answer to the enquiry will never be reached, 
for undue difficulty in adopting the prescriptions, and/or failure to generate satisfac-
tory dialogue, would indicate merely defects in the system used, rather than the fail-
ure of DGT as such. Whilst this is true in general, and may represent a difficulty 
endemic to research in educational technology, the criticism misses the point in the 
current context, where the aim is precisely to establish the suitability or otherwise of 
DC, as a prima facie reasonable system, with precedents for its usage. A negative 
answer would therefore be significant, though it would remain possible for another 
system to be found suitable. 

7. Empirical work concerning DC 

7.1 DC as a practical dialogue model 

The issue of the extent to which and the readiness with which would-be dialogue par-
ticipants are able to understand and put into practice DC has been addressed by an 
experiment, to be referred to as the "DC experiment", which was conducted in three 



Computational Uses of Philosophical Dialogue Theories 143 

stages-training, dialogue and debrief. The training session involved (i) explaining 
the rules of the game, (ii) holding and taping a brief (verbal) practice dialogue be-
tween experimenter and subject, and (iii) debriefing this session. The dialogue itself 
was held using remote terminals, in an attempt to simulate to some degree the envi-
ronment in which the dialogue system might eventually be used. Each subject was 
allotted, on an arbitrary basis, one of the propositions "capital punishment should be 
re-introduced" or "capital punishment should not be re-introduced", and told to try to 
convert their partner to this position. Subjects were told that their partner held the 
contrary position, and would be attempting a similar conversion. It was stressed that, 
whilst their position might not reflect their genuinely held belief, they should never-
theless genuinely seek to convert their partner. The debrief took place immediately 
after the debate, and involved written answers to a short questionnaire containing open 
questions. Data arising from the DC experiment was analysed to seek to establish the 
extent to which, and the ease with which, users can adopt the prescriptive framework 
of DC. This was achieved by matching the data against four specific criteria. 

The first concerns user understanding of the DC move types. On the majority of 
occasions subjects were found to classify correctly the moves made, suggesting that 
their view of what they are doing is generally consonant with the model's. Thus a low 
error rate can be predicted if users were presented with a menu of move types in a 
computerised version of the game. Subjects appeared to have most difficulty with the 
challenge move type, the only other difficulty of any significance concerning the 
resolution demand, which was occasionally incorrectly used. As a result ofthese lo-
calised difficulties, the dialogue game interface eventually developed (Hartley and 
Hintze 1990, see Section 8, below) adopts a menu system involving the DC locution 
modifiers (e.g. "I assert that...", "why .... 7"), rather than a command line interface or a 
menu interface involving the move type names. Further, the system displays the name 
of the move-type corresponding to user selection, enabling the user to check the 
accuracy of selection before "despatching" the move. 

The second criterion concerns how far users can utilise the move types in accord-
ance with the dialogue rules. The data suggests an overall picture of relatively infre-
quent rule breaks. Since all cases represent subjects' first exposure to the DC system, 
this seems an encouraging result from the point of view of using DC for computer-
based dialogue. Further, analysis of the nature of the rule breaks suggests that most 
would be avoided in a computational context by proffering to the user only legally 
available move types and contents, thus enforcing legality on the user. This would 
further reduce the already limited incidence of rule transgressions. 

The third criterion concerns the extent of subjects' appreciation of the effects of 
moves on their commitment stores. Subjects were not required to keep a running 
record of their commitment stores (CS), but rather to indicate as an answer to a de-
brief question what they believed their own and their partner's CS to contain. The 
evidence suggests that participants have major difficulty recalling the contents of 
their respective stores. No subject accurately reported the state of the CS, and some 



144 David Moore & Dave Hobbes 

were explicitly concerned about the difficulty of CS retention. It needs to be stressed, 
however, that the store is intended as a public and inspectable record of commitments 
rather than a memory test. This fact, together with the cognitive difficulties of un-
aided CS recall brought to light by the DC experiment, led to the eventual incorpora-
tion of a CS "window" into the DC interface, displaying the respective participants' 
stores. 

The fourth criterion concerns the extent of any learning difficulties. This is a cru-
cially important criterion, since were the system to impose too great a cognitive bur-
den on the participants, it would hinder rather than facilitate the debate, and thus achieve 
the opposite effect to the one intended. Some instances can in fact be found ofpossi-
ble learning difficulties on the part ofthe subjects. A number of concerns were voiced 
by subjects in connection with mastery of DC's rule set, suggesting that it would be 
hard to remember the rules. On the other hand, the generally successful manner in 
which subjects used the system, as indicated by the discussion of criteria 1 and 2, may 
be taken as supporting the notion that the system is learnable within a relatively short 
time scale, and the DC interface (Section 8) attempts to ameliorate any learning dif-
ficulties by giving an explanation of the nature of any rule transgressions, and making 
a copy ofthe rules available on-line. 

In sum, the data has suggested that people are generally successful in adopting the 
DC framework, and that the cognitive load it imposes is relatively light. Subjects have 
understood the move types well, rule breaks have been infrequent, and many of those 
that did occur could have been avoided with the aid of computational administration 
of the debate. Although there is some evidence of difficulty of mastering the DC 
framework, the cognitive load imposed by the need for such mastery would be con-
siderably diminished via computational management of the dialogue framework, and 
seems unlikely in any event to be such as to stand in the way of the production of 
educationally valuable dialogue. Acceptance of this last point is contingent upon an 
evaluation ofthe dialogue produced by the use of DC. This issue will now be consid-
ered. 

7.2 Qualitative evaluation of DC dialogues 

It is necessary, then, to conduct a qualitative assessment ofthe dialogue generated by 
the DC experiment. A further requirement is a comparative study involving the DC 
dialogues and dialogues generated by participants operating without DC-imposed con-
straints. For it may be that the attempt to exploit the formal properties of the DC 
framework causes unacceptable alterations, such as undue impoverishment, to nor-
mal conversation. Bench-Capon et al (1991) argue, for example, that because DC is a 
very formal system, adherence to its rules may result in a "somewhat stilted dialogue", 
and Mackenzie himself suggests that none of his games is "adequate" for "real life" 
argumentation (1990). An experiment was therefore conducted in which subjects de-
bated the issue of capital punishment without regard for (or indeed knowledge of) the 
strictures of DC (Moore 1993). 



Computational Uses of Philosophical Dialogue Theories 145 

A set of evaluative criteria were developed, and the dialogues resulting from the 
DC experiment and the unconstrained dialogue experiment were analysed against each 
(Moore 1993). In essence, the analysis suggests that the DC dialogues can be re-
garded as comparing favourably with unconstrained dialogue in terms of clarity of 
intention and commitment, coverage of issues and absence of fallacious argument. 
The analysis does, however, suggest two major concerns with the DC dialogues: the 
way questions are handled by the DC framework, and the initiative patterns that can 
result from adoption of the framework. 

DC permits only bipolar questions such as "is it the case that P?" and insists that 
the answer be one of"P," "not-P" or "no commitment to P." That the restricted range 
of question types may be problematic is suggested by the use of "illegal" question 
types in the protocols from both DC and unconstrained dialogues. Further, ineligibil-
ity of qualified answers to questions was seen as a deficiency by some subjects in the 
DC experiment. Mackenzie's acknowledgement that the question locution is crude 
(e.g. Mackenzie 1979b) appears therefore to be supported by the evidence. 

The problem may not be as great as it at first sight appears, however. In terms of 
question response, one simply replies in line with one's actual or potential commit-
ments; if one is not prepared for the commitment, the no commitment option should 
be chosen. It might be felt that this latter option would short-change both participants, 
in that having to use no commitment may make it harder for answerer to make his 
point, and harder for questioner to find out answerer's views on it. The latter problem, 
however, can be circumvented via challenge, or a statement to the effect that "I do not 
understand why you cannot commit yourself'. The former problem is not so readily 
soluble, in that the ability to expand one's view is contingent upon getting the initia-
tive and/or being challenged on it; the absence of any challenge, however, would 
suggest that respondem would be placed under no disadvantage by the non~committal 
answer. Similarly, ifrespondent wants to respond "P because Q" to the question "is it 
the case that P?", he should respond "P" and wait for a challenge to give his grounds 
for P (in this case Q). Absence of a challenge would suggest no strategic impoverish-
ment by not being able to give such grounds. 

Regarding the range of question types available, it could be argued that the only 
questions that have the right to a place in a debate as such are questions about the 
commitments of the participants-what they are already committed to or are pre-
pared to commit themselves to. And this is precisely what the DC question type al-
lows. All questions dealing with the "outside world", e.g. the comparative number of 
murders in UK and USA, are issues outside the debate per se-the debate is sus-
pended while the truth of the empirical matter is investigated. In a similar way, in a 
computerised system, the debate may be suspended while "note cards" on the ma-
chine are consulted. Further, although the experiment with unconstrained dialogue 
contained a number of non-DC question types, it turns out that most can be "trans-
lated" into an equivalent DC-acceptable form (cf. Moore 1993). The restriction to 
bipolar questions need not therefore, in a computer context, lead to impoverishment. 



146 David Moore & Dave Hobbes 

The second major concern with the DC dialogues revealed by the experiments 
involves the presence of relatively stable initiative patterns. The tendency was for one 
subject continuously to ask questions or make challenges, with his interlocutor able 
only to respond. The evidence suggests, however, that in a computational environ-
ment, initiative will largely reside with the student, and on Girle's argument (1986) 
this may be seen as a good thing, allowing the student to probe critically for informa-
tion rather than act as a passive receiver. Further, heuristics can be devised that will 
prevent the computer from unreasonably retaining the initiative, on those occasions 
when it no longer resides with the student. 

In sum, the qualitative assessment of debates generated via the dialectical system 
DC suggests that the system can be regarded as a valid prescriptive framework for 
useful educational dialogue. The comparison with dialogue generated without any DC-
imposed restrictions suggests that the qualitative differences between the dialogues 
produced by the two paradigms are in practice relatively little: most of the apparent 
restrictions in DC can be overcome, and DC avoids some difficulties that can beset 
unconstrained dialogues. 

7.3 Strategy in DC 

DC provides a regulatory framework within which dialogue participants must operate, 
but in common with other such frameworks, it allows room for participant discretion 
as to the precise operation ofthe system. In the case of DC, what might be considered 
its main asset, the simplicity engendered by its relatively slender rule set, may also be 
its downfall, for the cost of the simplicity is a reliance on the strategic wisdom ofthe 
participants, for example to maintain relevance. Further, the avoidance of unscrupu-
lous yet legal play, e.g. constantly retracting and reasserting a proposition, must also 
rely on the strategic wisdom ofthe participants (cf. Mackenzie 1980). 

Since the results from the previous section suggest that the generated dialogues 
can be considered qualitatively acceptable, this reliance on strategy seems justified. 
From the computational point of view, however, an understanding of the strategies 
adopted becomes vital. The computer will also follow DC's slender rule set, and it 
will be equally reliant therefore upon appropriate strategic wisdom. Indeed, the abil-
ity to capture this wisdom could be seen as the crucial aspect of the current research. 
F or in this strategic knowledge lies the potential for a link between the logic of argu-
mentation, as modelled by DC, and the semantics of discourse. And without this link, 
it might be suggested, DC will add nothing to currently available computer dialogue 
systems. It could be argued, against this suggestion, that DC would potentially be of 
value even without such a link. For it is possible for locutions to be syntactically 
correct and semantically meaningful, yet inappropriate at the pragmatic level. Were 
DC to provide a reasonable framework for debate at this pragmatic level, it would still 
have merit, even though it might in itself give no guidance in regard ofthe semantic 
content of the constituent moves in the debate. Nevertheless, the case for the adop-
tion of DC would be considerably strengthened were a link to the semantics to exist 



Computational Uses o/Philosophical Dialogue Theories 147 

or to be forged. For the upshot of the argument would be (at best) that DC is a prag-
matic controller awaiting development of a suitable semantic processor to provide 
things for it to control. This would be cold comfort to anyone awaiting a working 
debating program. Further, one possible weakness of DC is that the system is largely 
neutral vis-a-vis focus and content considerations. That this has not turned out to be a 
problem has been seen as a result of participants' strategic reasoning, and this reli-
ance on strategy again involves a link to discourse semantics. 

It should be noted that any link to the discourse semantics forged via such strate-
gic decisions is largely an extra-game consideration: all DC does is legitimise a set 
of move types given the prevailing circumstances, and occasionally give some indica-
tion of the semantic possibilities. One might think that DC would indicate what is 
legally available, a strategic choice would be made from this range, and the choice 
would then be 'filled up' with semantic content. Such a view would be too simple, 
however, for the strategic decision is likely to rely heavily on the available content, 
and a strategic decision may be needed between alternative contenders for the content 
(different supporting evidence, for example). In a computerised system, some inter-
play between strategic and semantic components may therefore be needed. The task 
is, therefore, to analyse the DC data with a view to deriving guidelines (heuristics) 
which will help a move maker select a sensible move. The formalisation attempt can 
be expected to reveal the nature of the strategy-semantic interplay, and thus to move 
towards a link between DC and discourse semantics, and hence towards computa-
tional dialogue participation. 

Analysis of the game's properties makes it clear that at any stage in the game each 
participant has a double choice to make, concerning move type and semantic content. 
Further, at any stage in the game, one participant must be regarded as "having the 
initiative", in that no constraints are placed upon his choice of move type and content 
by his partner's previous move. At any given stage ofthe game the initiative must rest 
with one or other of the participants. The strategic problem is simpler for the partici-
pant who has not got the initiative. For choice of move type is more constrained-a 
question must be answered, a challenge must be responded to by support, withdrawal, 
or (occasionally) resolution demand, and a resolution demand must be resolved. The 
content issue is heavily constrained following a question, concerns which statement 
to withdraw following a resolution demand, and is (in principle) totally unconstrained 
following a challenge. When a participant has the initiative, his strategic options are 
wider, since he has no constraints on move type and very few on content. There are 
three conditions in which a participant has the initiative: beginning of game, partner 
has used the No Commitment move type, partner has made a statement. 

Now, it can be suggested that there are decisions at three levels, some or all of 
which are involved in a participant's strategic decision as to his next move. At/evel 1 
the issue is whether (a) to retain the focus or (b) to change it. "Retain the focus" can 
be taken to mean continuing the same line of argument, which will involve continuing 
to attempt to substantiate or undermine a particular proposition. A decision at level 1 



148 David Moore & Dave Hobbes 

is needed only where participant I (call him "P") has the initiative, since otherwise 
the game rules impose on P the obligation to address his partner's previous move 
(call P's partner "0"). Where a decision is needed, one might intuitively expect there 
to be a presumption in favour of addressing partner's previous utterance (and there-
fore retaining the focus). In general, were the presumption in favour of addressing 
partner's previous utterance not to hold, conversations would be highly disjointed. 
And empirical analysis of the DC data shows that where the maker ofthe move has the 
initiative, he opts to continue the line of argument in the majority of cases, and usu-
ally to address directly his partner's previous move. Indeed, the decision at this level 
will often be pre-empted by the heuristic adopted at the previous turn. For example if 
the subject previously challenged with a view to rebutting arguments, he is inevitably 
likely to want to address the forthcoming reasons. 

At level 2 there are again two alternatives for P. He can seek to demolish O's 
position, that is seek to have him remove from his CS propositions which he has used 
to support his thesis. This may be done either by showing a contradiction between, or 
an unacceptable consequence of, his CS (this may initially involve adding to O's store), 
or by demonstrating that he has no reasonable support for propositions in his CS. The 
ultimate aim is to remove all support ofO's position. Alternatively, he can build his 
own position, by making statements the acceptance of which, or asking bipolar ques-
tions the answers to which, ultimately imply the truth of his thesis. Level 2, like level 
I, will be by-passed ifP does not have the dialogue initiative. Further, the level 2 
decision is needed only at the beginning of the game, and on those occasions when the 
level I decision is to change focus. As long as the level I decision retains the current 
line of argument, the decision between build or demolish made at the beginning of 
that line will continue. Empirical analysis of the DC data yields insufficient evidence 
to suggest heuristics re the beginning of games. Similarly, the evidence with regard to 
level 2 decisions after focus shifts is somewhat conflicting, in that the protocols 
contain evidence both of subjects desiring to build up stocks of commitments, and of 
subjects desiring to demolish their opponent's support of his thesis. There appears 
therefore to be no clear preference for either option at level 2, implying that the 
computer would be on safe ground whichever it chose. 

At level 3 the decision involves which method to adopt in fulfilment ofthe objec-
tive set at levels I and 2. IfP is building he may form a plan, consisting essentially of 
a set of propositions he needs to add to O's CS, which ultimately combine to imply 
the truth ofP's own thesis; similar plans, targeted at propositions 0 is using to sup-
port his thesis, could be used in demolition attempts. The plan can be activated at level 
3 either by stating the first proposition in the plan, or posing a bipolar question with 
the proposition as its content. If the response is the one hoped for (e.g. 0 replies in 
the affirmative), P can be expected to continue on to the next proposition in the plan 
(a level I decision). If the response is not what was hoped (e.g. 0 answers in the 
negative, or objects to the proposition), some level 3 mechanism will be needed for 
dealing with this response, assuming that the decision at levell is to continue with the 
line of argument. 



Computational Uses of Philosophical Dialogue Theories 149 

The following heuristics for level 3 have been derived from the evidence: 

if a question has been asked: 
answer in such a way as to avoid unwelcome commitments; 
give answers for which good reasons can be given; 
answer truthfully. 

if a challenge has been made: 
here one would normally reply with some support, but it is not clear from 
evidence so far collected how particular supports are chosen. 

if a resolution demand has been made: 
(tentatively proposed): withdraw whichever conjunct is the least harmful. 

if a no commitment move has been made: 
in the event of partner withdrawing a proposition previously used in sup-
port of his thesis, either check whether partner retains adherence to the 
thesis, and challenge for further reasons, or switch focus; 
(tentatively proposed): in the event of withdrawal ofa proposition at a lower 
level of a support hierarchy, seek withdrawal of propositions supported by 
that proposition. 

if a statement has been made: 
if there is (an expressed) contradiction in partner's CS, resolve it; 
ifthere is any evidence directly contradicting any of partner's statements, 
state it; 
seek substantive objection to partner's commitments, and pose questions 
with a view to making partner accept that objection; 
seek out by challenge partner's arguments, with a view ultimately to rebut-
tingthem. 

Note that the heuristics for dealing with a statement are listed here in the pre-
ferred order of choice. That is, if a has commitments which P wishes to dispute, the 
first heuristic would be the strongest method of attack for P, the fourth the weakest. 
The fourth might be regarded as an attack in itself, or as a means of elucidating a's 
reasoning with a view to further, more direct attack (via one of the other heuristics). 

8. DC as a computational model 

A user interface has been developed, by Hintze (Hartley and Hintze 1990), that al-
lows two players to engage in a DC dialogue supported by a computer-based gameboard 
and referee. An example of the system ready for use is shown in Figure 1. (next page). 
The interface allows a user to select, from a menu, a move type from the range al-
lowed by DC. The user then types in the content of their move. The system has no 
knowledge ofthe semantics of the players' moves, but is able to compute the logical 
connections between statements made, and thus decide what must go in each player's 
commitment store, and rule as to whether the move is legal. Commitment stores are 
displayed in "commitment set" windows. Participants are thus relieved ofthe cogni-



150 David Moore & Dave Hobbes 



Computational Uses of Philosophical Dialogue Theories 151 

tive burden of commitment store recall. The interface is able to referee all moves 
made according to DC's dialogue rules, thus ensuring a legal debate. 

Certain educational advantages of the use ofthis dialogue game interface can be 
suggested. It could be used as a spur to develop arguments logically, to remain con-
sistent, and to improve debating skills. Using the interface with different dialogue 
partners forces application of these skills to information offered by the other player, 
thus potentially generating new insights into the domain of enquiry. The system offers 
the facility to record dialogues, and to replay previous dialogues either in full or to 
some specified point, from which users may continue to add their own contributions. 
Thus dialogues from different participants can be studied, and alternative lines of 
thought experimented with. 

Indeed, the successful implementation ofthe DC interface is all that is required to 
demonstrate the computational tractability of DC as a dialogue model. For DC is seen 
as a framework through which two people can engage in dialogue with each other, by 
adhering to certain rules, and by being made accountable, via their commitment stores, 
for what they say and accept during the debate. And the DC interface provides pre-
cisely such a vehicle. It thus provides a computerised version of the DC system, and 
demonstrates DC's computational tractability. Of more educational interest, however, 
is the possibility of having the computer act as aparticipant in the dialogue. This will 
involve enabling the computer to make its own contributions, as well as acting as 
referee and administrator of the debate. 

Such a system will be required to take part in, and referee, debates with individual 
students. The computer (C) would therefore seek the student's (S's) opinion on a 
controversial subject, e.g. capital punishment, inform the student of its own (oppos-
ing) view, and invite S to participate in debate on that subject. During play C will need 
to recognise the type and content ofS's moves, and to rule as to their legality. Assum-
ing the move is deemed legal, the commitment stores of both Sand C may need to be 
updated, and the dialogue history updated. All these requirements are met by the cur-
rent DC interface. Next, C needs to select a suitable move type and content with 
which to respond to S's move, and to arrange for the move to be displayed to S. S in 
tum makes his response, causing the cycle to start again, until one or other party 
either "wins", or withdraws through lack oftime or interest. A software architecture 
for a system to fulfil these requirements has been designed, and is the subject of on-
going research and development work (Moore 1993). 

One difficulty with the proposed system concerns knowledge representation, widely 
seen as the central issue in AI in general (e.g., Bench-Capon 1990), and ITS in particu-
lar (e.g., Yazdani 1986). In the current context the challenge is to represent the sys-
tem's domain knowledge in such a way as to enable provision of statements that can be 
used in response to questions, and of statements that can be used to support other 
statements, Satisfaction of these two requirements will provide the knowledge to en-
able C to operationalise the strategic heuristics discussed earlier. For example, demo-
lition attempts can be catered for by negating the proposition which is the target of 



152 David Moore & Dave Hobbes 

the demolition, and recursively seeking to support that negation. Building attempts 
can be handled by recursively supporting the thesis itself. Starting at the "bottom" 
(i.e. with the last proposition to be returned) would yield a "distancing" strategy; the 
first support statement generated would form the strategy's "comer" (Walton 1984). 
Both are analogous to Cohen's (1987) pre-order form of argument. IfC is subject to 
a challenge from S, it can respond with the first statement found in support of the 
challenged statement; in effect this resembles Cohen's post-order form of argument. 
If the proffered support is itself challenged, C can use the next in the recursive line of 
support. An alternative method of responding to challenge would be recursively to 
apply the support relationship, and offer as support the last statement found; this 
would in effect leave many enthymematic premises, which S could uncover by further 
challenges. 

One way of engineering the knowledge base to fulfil these requirements is pro-
vided by Bench-Capon et al. (1990). They utilise a modified version ofToulmin's 
argument formalism to construct a knowledge base which can be used by an inter-
preter to generate, on-line, paragraphs oftext. In the current context the idea would 
be to use the output from the interpreter as the basis for responses to user challenges, 
using the appropriate "data" or "warrants" relating to the claim that has been chal-
lenged. Such an arrangement would enable a generic knowledge base to be held, and is 
discussed further in (Moore 1993). 

Another interesting approach to the issue of retrieving appropriate information 
from the knowledge base is to draw upon work using "rhetorical predicates" (Grimes 
1975) to generate coherent text at the paragraph level or beyond (Pilkington 1992b). 
Rhetorical predicates have been used successfully in a number of computer-based 
implementations (see, e.g., Tattersall 1993, Maybury 1992). The knowledge-base can 
be represented in a generic or "pure" manner (Pilkington 1992b), with predicates 
abstracting out the appropriate content for a particular utterance (Maybury 1992). 
This would enable different domains to be accessed by the debating system, thus pro-
viding debates across a range of issues. An extended example of the use of rhetorical 
predicates in a dialogue game context is provided in Pilkington et al. (1992). 

The use of rhetorical predicates also offers a potential solution to a second major 
difficulty with the proposed system, the handling of substantive student input. Com-
puter processing of free range natural language input is a large problem (e.g. Allen 
1987), beyond the scope of the current project. It may be possible, however, to use 
rhetorical predicates as an intermediary language, forming a level between the sys-
tern's semantic domain representation and the user interface. The idea is that the predi-
cates are iconised and the student uses them in the cut-and-paste construction of his 
substantive moves. In this way the student can construct relationships between ob-
jects, to form the propositional content of his moves. He mf-Y, for example, select a 
cause-consequence predicate, and objects "angina" and "fatigue", to suggest that an-
gina can give rise to fatigue. In this way the system can provide a range of semantic 
links which can give rise to a common understanding between system and user. The 



Computational Uses of Philosophical Dialogue Theories 153 

system can then test for consistency between relationships and objects proffered by 
the user, and existing relationships and objects in the domain knowledge base, and 
decide, on this basis, whether to accept the statement or to seek to demolish it via the 
strategic heuristics. Indeed, it may be that in certain circumstances, the domain knowl-
edge base itself should be altered as a result of relationships suggested by users, so 
that a learning element can be incorporated into the system (cf. Pilkington et at. 1992). 

9. An example 

An attempt will now be made to give a brief overview of a conceptualised system 
operating with the topic of debate concerning whether all dogs should be extermi-
nated. The discussion here will assume the idea of using hypertext-like cards, with 
links based on Bench-Capon et at.'s argument schema. The discussion should both 
illuminate the workings of the card-based approach, and the ability of the DC frame-
work and strategic heuristics to operate to produce debate. 

The card-based approach assumes that a set of inter-linked cards can be created. 
This may be done by the system designer, or by some domain "expert", or, in a school 
setting, by a member ofthe student body, since the creation of the resulting concep-
tual map might itself be of educational benefit. A starting point is likely to be the 
creation, perhaps by a brainstorming session, of a number of (possibly) relevant propo-
sitions. In this case suitable propositions might be: 

dogs protect people 
dogs help people 
dogs give amusement 
dogs foul pavements 
dogs bite people 
dogs use up resources 
dogs can be guide-dogs for the blind 
dogs can help round up sheep 
dogs can guard property 
dogs are a popular pet 
making dog food provides employment 
unwanted dogs are turned loose in the street 
dogs must be registered 
people are cruel to dogs 
dogs eat food 
people can be prosecuted for not looking after dogs 
RSPCA looks after dogs' interests 

Having collected a prelimin¥y set of propositions, these need to be linked in 
accordance with the model of argumentation being adopted. In this instance a simpli-
fied version of the Bench-Capon et al. model will be used, in which only the evidence 
and rebuttal links are used. The latter will be treated as a straightforward candidate 



i;3 
..(;) 
,.(;) 

~ 
<:U 
;:. 

~ 
~ 

~ 
~ 
~ 

~ 
'<:t 
Vl ...... 

(dOs-------- ~ 
~s should not be extenninate:l_"_ 

er ev ev ..... 
.... 

IdOSSprotect -] IdogS help IdOgs give pleasure 

reb 

.... 

.... .... 
....eV 

m"~OgS should be extenninated 

ev ev ev 

ev ev ev ev lcause disease luse up resources] EJ Iloose in street] 

19uard dogs! [guide dogsllsheep dOgs! !most popular pet I 
reb 

KEY 
ev - evidence 
reb - rebuttal reb 

ev 

f~~ 

IRSPCA protects dogs' interests ] 

ev reb 

[~i food I 

Figure 2 • Conceptual Map of the Dog Domain 

reb 

lr;tbe I r 'stered 



Computational Uses of Philosophical Dia/ogue Theories 155 

"objection", since the "unless" construct, for which it was originally intended, is not 
available in DC. 

There are many ways in which the propositions can be linked. This is a necessary 
condition of, and part ofthereason for, debate in the first place; people with identical 
conceptual maps cannot have a meaningful debate. The representation in Figure 2 
(opposite) is, therefore, a suggested representation only. Given this conceptual map 
of the dog domain, an analysis of how a computer (C) might hold a debate with a 
student (S) can be presented. The following debate will be traced on a line-by-line 
basis. 

1 C Is it the case that dogs should be exterminated? (S 1 ?) 
2 S NotSl. 
3 C Dogs should be exterminated. (S 1) 
4 S WhySl? 
5 C Dogs cause disease. (S2) 
6 S WhyS2? 
7 C Dogs foul the pavement. (S3) 
8 8 Why S3-> 82? 
9 C No commitment to S3 -> S2. 
10 S Why82? 
11 C No commitment to S2. 
12 8 Dogs give pleasure to people. (84) 
13 C WhyS4? 

14 S Dogs are the most popular pet. (85) 
15 C People are often cruel to dogs. (S6) 
16 S The RSPCA protects dogs' interests. (87) 
17 C Is it the case that dogs eat food? (S8?) 
18 S S8. 
19 C Is it the case that dogs use up the world's resources? (89?) 
20 S 89. 
21 C Is it the case that 89 -> dogs should be exterminated? 
22 8 Not (89 -> dogs should be exterminated). 
23 C' Why? 
24 S Making of dog food provides employment. 

Lines 1-3 can be regarded as setting up the game's preconditions, with C taking, at 
line 3, the opposing thesis to 8. At line 48 challenges C's thesis. This is done by 
selecting "Why ... " from the move menu in the DC interface, and 81 from the state-
ment record. This suffices to tell C that a challenge has been made, and the heuristics 
suggest that a statement is required in response. C therefore searches the cards for 
any which have an "evidence" link to the challenged proposition; any such will be a 
candidate for the statement to be made. In this case there are four such candidates 
("dogs cause disease", "dogs bite people", "dogs are often loose in the street", "dogs 
use up the world's resources"), but it can be assumed that C will use the first appropri-



156 David Moore & Dave Hobbes 

ate card it finds, and that in this case the "winner" is "dogs cause disease". This is 
therefore posted as C's move (line 5). Identical reasoning causes C's response in line 
7 to S's challenge at line 6. At line 8, S takes the option of challenging the implication 
concomitant upon C's move at line 7. C again searches the card knowledge-base, but 
finds no card with an evidence link. The only available option therefore is to withdraw 
the challenged statement (line 9). Since there is no longer any support for S2, this is 
again challenged (line 10); a search of the cards shows no suitable evidence link, so C 
must now withdraw S2 (line II). 

S now starts a build attempt (line 12). This is done by selecting "I assert that..." 
from the move menu, and "dogs give pleasure" from the card-stack. This stack should 
be available to S in a separate "domain" window; the encoded links between the cards 
should not be shown to S, since this would detract from the requirement imposed on 
S to build his own argument given the available information. C is therefore aware that 
a statement has been made. All three levels of strategic decision making are therefore 
applicable. Since there is a presumption at level 1 in favour of addressing the previous 
utterance, and since there is nothing here to suggest otherwise, C will address S's 
utterance. Since there is an evidence link from S's chosen card to S's thesis, this level 
1 decision implies in this case that the level 2 decision must be in favour of demoli-
tion, and that the preferred order of heuristics is: 

(a) If there is (an expressed) contradiction in S's commitment store, resolve. 
(b) Ifthere is any evidence directly contradicting any ofS's commitments, 

then state it. 
( c) Seek substantive objection to any of S' s commitments, and pose questions 

with a view to making S accept that objection. 
(d) Seek out (by challenge) S' s arguments, with a view ultimately to rebutting 

them. 
Examination ofS's commitment store reveals that (a) is not available (the store 

contains no contradiction). The only statement in S's store, apart from his thesis, is 
the one just made, and since there is no card with a "rebuttal" link to this statement, 
heuristic (b) must be ruled out. Similarly, there is no evidence for the negation ofS's 
statement, so that heuristic (c) also fails. Heuristic (d) must therefore be utilised 
(line 13). S replies by suggesting that dogs are the most popular pet (line 14). 

S's selection of the statement move type implies that all three levels of strategy 
again need to be considered. The default condition at level 1 applies (since S has 
responded to C's challenge), and level 2 is therefore redundant. C therefore needs to 
consider the same set of heuristics as on its previous move. Heuristic (a) is again 
unavailable, but since a rebuttal is present in the card knowledge base, heuristic (b) 
does fire, and C makes a statement of the rebuttal (line 15). S finds a card to use as 
rebuttal ofthe rebuttal (line 16). 

C needs to respond to this statement. The knowledge-base contains no rebuttal to 
S's statement, and no evidence for the statement's contradiction, and since the use of 
a challenge would lead to either a complex line of argument or a non-committal re-



Computational Uses of Philosophical Dialogue Theories 157 

sponse, the level 1 decision is to change focus. Given an assumption in favour of 
building, G needs to seek suitable support of its own thesis ("dogs should be extermi-
nated"). This can be done by seeking cards with evidence links to this thesis. Assume 
that C will adopt the first (unused) such card, and that in this instance the card "dogs 
use up the world's resources" is found. Evidence for this, in order to mount a distanc-
ing strategy, will also be sought, and so on recursively. In this instance the loop ends 
with the next card ("dogs eat food"). The level 3 decision therefore recommends 
posing this as a question (line 17).At line 18 S answers in the desired way, so the next 
stage in the distancing plan is selected, and the question put (line 19), to which S again 
acquiesces (line 20). C therefore asks what in effect would be the victory question 
(line 21), but does not get the response sought. The level 3 heuristics therefore fire 
against unwanted response, (d) is the only one applicable, and the challenge is made 
(line 23), to which S replies in line 24. 

This example (which could of course continue) shows that the idea of utilising a 
hypertext style approach based on Bench-Capon et al. 's argumentation model, can 
furnish appropriate content for the dialogue moves. Although not shown in the exam-
ple, the links between the cards could be used by C when electing how to respond to 
questions, resolution demands, and withdrawals for S. Further, the computer could 
have argued the case against the extermination of dogs, thus permitting debate to take 
place irrespective of whether the initial position ofS is for or against canine extermi-
nation. The example also shows that the rules and strategies are generic, in that a set 
of cards relating to a different domain could be used, with no requirement to alter the 
dialogue rules or strategic heuristics. 

The card-based approach also provides a short-term answer to the input problem. 
The contents of the cards can furnish S's substantive moves, and a blank card would be 
available to cater for S disenchantment with those on offer. Further, by tracing the 
links between cards selected by S, C can interpret them as being favourable or unfa-
vourable to its own position, and thus decide whether they form sensible objects of 
attack. A potential difficulty in this context might arise ifS uses cards in ways which 
run counter to the encoded links, using as support of his position, for example, cards 
which have evidence links to C's thesis. In this case the best that can be suggested, 
perhaps, is that C challenge the use of the card (i.e. challenge the implication of which 
it forms the antecedent). Substantive S response will necessitate abandoning the cur-
rent line of argument, and subsequent knowledge-base reconsideration by its "author". 
Similarly, S may not accept that a proposition (P) used by C as defence of a further 
proposition (Q), should rightfully be so used. The issue is again one of disagreement 
with the accuracy of the conceptual domain map. In such circumstances, S should 
challenge the proposition's status as evidence (as in line 8). C will not be able to 
defend itself, and must therefore withdraw the statement; S will not therefore be dis-
advantaged by any refusal to accept the conceptual map. In the longer term, the knowl-
edge base author may wish to strengthen the knowledge base by inserting (what is 
seen as) suitable support for the evidence relationship; this would be equivalent to 



158 David Moore & Dave Hobbes 

Bench-Capon et al. 's "warrants" (1990). Similar arguments apply to S's disagreement 
with alleged rebuttal relationships between propositions. 

Summarising, the analysis in this section has followed the workings of a concep-
tualised system as it seeks to utilise the strategic heuristics in order to participate in 
debate. Current work involves a developing a full implementation of the system. A 
problem implicit in the discussion concerns the notion of relevance. Following Carlson 
(1983), relevance is seen as a strategic concern, and it could be argued that, as far as 
the computer's contributions are concerned, relevance is provided by the heuristics, 
given appropriate service from knowledge base. It might also be contended that rel-
evance ofS input could be taken for granted, since C would assume rationality on S's 
part. A difficulty arises, however, in the case of genuine irrelevancies, for were C to 
fire the heuristics against these, rather than ignore them, the debate could spend some 
time in a wholly inappropriate direction. Nevertheless, since C has no means of dis-
tinguishing clever strategy from time-delaying or attempts to psyche out the machine, 
it will need to assume a co-operative user, and treat all inputs on merit. 

A co-operative user also needs to be assumed in connection with the resolution 
demand move type. For by leaving the list (V) of preferred valid argument schemata 
unspecified, and always withdrawing following a legal resolution demand, C is open 
to false accusations ofinconsistency, and, worse, may confuse the genuine student by 
withdrawing commitments which are not "really" inconsistent. This may be a problem 
in the short term, but is very unlikely to arise in practice, and, in the longer term, a 
specification for V can be arrived at, leaving C the possibility of an objection move 
where V is not appropriately invoked. The converse danger, that C can pose resolution 
demands concerning any statements in S's commitment store, can be avoided by al-
lowing usage of resolution demand only in the face of direct inconsistency (P, not-P), 
and inconsistency involving modus ponens (P -> Q, P, not-Q). 

10. Conclusions and further work 

The research outlined in this paper represents a preliminary attempt to investigate the 
potential of DOT as a vehicle for enhanced human-computer communication. The 
investigation has involved empirical tests ofthe suitability ofthe DC system, and an 
enquiry into the tractability of the system for computer-based applications. 

The overall conclusion from the research is that DOT would appear to offer a 
promising vehicle for computer-based dialogue. This has been demonstrated from 
both the computational and user viewpoint. From the computational point of view, the 
demonstration has been via discussion of an operational dialogue game interface, and 
the construction of a suitable architecture for, and program conceptualisation of, com-
putational participation in dialogue, utilising the simplicity and order imposed on dia-
logue via the game moves and rules, a set of heuristics, and previous successful work 
on argument schemata (cf. Bench-Capon et al. 1990) to access knowledge based sys-
tems. From the user's point of view, there is a trade-off, as in all computer appJica-



Computational Uses of Philosophical Dialogue Theories 159 

tions, between the difficulty of adopting the system's strictures, and the advantages to 
be gained by so doing. Adapting to DC's regimen has been shown to impose a rela-
tively light cognitive load on users, and the computer interface liberates players from 
many of the game's potential complexities. Given this, and the advantages ofthe en-
hanced dialogue that DC allows, the trade-off seems favourable: the potential gain is 
sufficient t.o merit the effort required to adopt the system. In brief, it has been shown 
that DC provides the tightness of dialogue needed for computational tractability, whilst 
still being advantageous from the user's point of view. 

There are a variety of interesting ways in which the research can be carried for-
ward. In the short term, the aim will be to develop further the debating system, to yield 
an application which adopts a position diametrically opposed to that of the student, 
and engages him in debate. Interesting enhancements to this "basic" debating system 
can be suggested. One is to extend the system to use interactive video and other mul-
timedia technology (Hobbs and Moore 1994). One use for video might be as an initial 
stimulus to encourage students to enter debate, showing, for example, a documentary 
about abortion, or footage of philosophers in discussion. The computer could also 
use, where appropriate, video as its contribution to the debate, in the manner of DE-
CIDER (Bloch and Farrell 1988). Interesting questions would then arise as to how to 
select the best form of output (cf. DuBoulay and Sloman 1988). A further advantage 
of multimedia technology would be to provide, via hypermedia links highlighted in 
the computer's output, opportunities for the student to clarify points and/or terminol-
ogy he does not understand (Bielawski and Lewand 1991). Such a hypermedia facility 
could be expected to ease the lack of a clarification move in DC. Such developments 
suggest that the debating system could eventually find a place in an overall "study 
station" (Tait 1987), with the debate and other facilities operating to each others' (and 
therefore, of course, to the student's) mutual advantage. 

A long term goal will be to move beyond debate, into other forms of dialogue 
interaction. The ultimate goal would be a system capable of selecting and engaging in 
a variety ofinteraction styles with students, according to the demands of the subject 
matter and the educational interests of the individual student. It is clear that many 
interesting and challenging issues need to be addressed, concerning in particular the 
development of suitable dialogue models, and computational utilisations thereof, be-
fore this goal can be reached. However, this paper hopes to help bring together two 
groups of workers (philosophers of dialogue and designers of ITS) so that each can 
learn from, and contribute to, the experiences of the other, and thus move towards 
achieving that goal. 

Acknowledgement 
The authors gratefully acknowiedge Professor 1. Roger Hartley, Dr. Rachel Pilkington, and David 
Hintze, for many invaluable discussions in connection with this work. 



160 DavidMoore & Dave Hobbes 

References 

Allen, J. (1987). Natural Language Understanding; Benjamin/Cummings. 
Anastasi, A. (1982). Psychological Testing. MacMillan. 
Bailey, J.1. (1989). Explanation-Giving Systemsfor Learning Decision-Making Skills. PhD. 

Thesis, Computer Based Learning Unit, School of Education, University of Leeds. 
Baker, M. (1989). A model for tutorial dialogues based on critical argument. In D. Bierman, J. 

Breuker and J. Sandberg, (eds.),ArtificiaIIntelligence and Education. Proceedings of the 
Fourth International Conference on Al and Education: 24-26. 

Baker, M. (1994). A Model for Negotiation in Teaching-Learning Dialogues. Journal of Artifi-
cialIntelligence in Education. vol. 5, no. 2: 199-254 . 

Barth, E.M. (1982). A Normative-Pragmatical Foundation of The Rules of Some Systems of 
F ormal

3 
Dialectics. In Barth and Martens (1982). 

Barth, E.M. and Krabbe, E.C.W. (1982). From Axiom to Dialogue: A Philosophical Study 
of Logics and Argumentation. Walter de Gruyter. 

Barth, E.M.and Martens, J.L. (Eds.). (1982). Argumentation: Approaches to Theory Forma-
tion. JohriBenjamins, B.V. 

Bench-Capon, T.J .M., Lowes, D .and McEnery, A.M. (1990). Using Toulmin' s Argument F or-
malism to Explain Logic Programs. Proceedings Explanations Workshop V. Manchester. 

Bench-Capon, TJ .M. (1990). Knowledge Representation. Academic Press. 
Bench-Capon, T.1. M., Dunne, P.E.S. and Leng, P.H. (1991). Interacting with Knowledge Based 

Systems through Dialogue Games. Proceedings of Eleventh International Conference. 
Expert Systems and their Applications, vol. I. Avignon, May 1991. 

Bielawski and Lewand. (1991 ).Intelligent Systems Design. Wiley. 
Bloch, G., Farrell, R. (1988). Promoting Creativity Through Argumentation; DECIDER: acase 

based teaching system.ITS-88: 243-249. 
Booth, P. (1989). An Introduction to Human-Computer Interaction. Lawrence Erlbaum. 
du Boulay, B.and Sloman, A. (1988). Bread Today, Jam Tomorrow: The Impact of AI on Edu-

cation. Proceedings of Fifth International Conference on Technology and Education. 
vol. I: 82-85. 

Carlson, L. (1983). Dialogue Games: An Approach to Discourse Analysis. D. Reidel. 
Carlson, L. (1984). "Well" in Dialogue Games. John Benjamin. 
Carlson, L. (1984b). Focus and Dialogue Games. In L. Vainaand 1. Hintikka(eds.), Cognitive 

Constraints on Communication. pp. 295-333. DReidel. 
Cawsey, A. (1990). A Computational Model ofExplanatory Discourse: Local Interactions in a 

Plan-Based Explanation. In N. Luff, (ed.), Computers and Conversation. Academic Press. 
Clark, H.H., Schaefer, E.F. (1989). Contributing to Discourse. Cognitive Science 13: 259-294. 
Cohen, R. (1987). Analysing the Structure of Argumentative Discourse. Computational lin-

guistics, vol. 13 numbers 1-2: 11-24. 
Eisom-Cook, M. (1985). Towards a Framework for Human-Computer Discourse. In P. Johnson 

and S. Cook, (eds.), People and Computers: Designing the Interface. Cambridge Univer-
sity Press. 



Computational Uses of Philosophical Dialogue Theories 161 

Elsom-Cook, M. (1988). Guided discovery learning and bounded user modelling. In Self (i988). 
Finkelstein, A. and Fuks, H. (1990). Conversation Analysis and Specification. In N. Luff, (ed.), 

Computers and Conversation. Academic Press. 
Finkelstein, A. and Fuks, H. (1989). Multi-Party Specification. Proceedings Fifth interna-

tional Workshop on Software Specification and Design. Pittsburgh. 
Frohlich, D.M.and Luff, P. (1990). Applying the Technology of Conversation to the Technology 

for Conversation. In N. Luff, (ed.), Computers and Conversation. Academic Press. 

Fuks, H" Ryan, M. and Sadler, M. (1989). Outline of a Commitment Logic for Legal Reasoning. 
Proceedings of Third International Conference on Logic, Informatics, Law. 

Galton, M. and Simon, B. (Eds.). (1980). Progress and Performance in the Primary Class-
room. ROJ,ltledge and Kegan Paul. 

Garrison, D.R. (1991). Critical Thinking and Adult Education: A Conceptual Model for Devel-
oping Critical Thinking in Adult Learners. International Journal of Lifelong Education, vol. 
10, no. 4: 287-304. 

Girle, R.A. (1986). Dialogue and Discourse. In G .Bishop and W . Van Lint, (eds.), Proceedings 
of the Fourth Annual Computer Assisted Learning in Tertiary Education Conference, 
Adelaide. Distributed by Office of Continuing Education, University of Adelaide. 

Gordon, T.G. (1994). The Pleadings Garne: An Exercise in Computational Dialectics. Artificial 
Intelligence and Law, vol. 2, no. 4: 239-292. 

Grimes, J.E. (1975). The Thread of Discourse. de Mouton. 
Harnblin, c.L. (1970). Fallacies. Methuen. 
Hamblin, C.L. (1971). Mathematical Models of Dialogue. Theoria, vol. 37, no. 2: 130-155. 
Hamblin, C.L. (1987). Imperatives. Basil Blackwell. 

Hartley, J.R. and Hintze, D. (1990). Dialogue and Leamer Modelling. In SA Cheri, (ed.), 
Student Model Acquisition in a Natural Laboratory (NATLAB); GEC DELTA Project D-
1016 Final Report, Brussels. 

Hartley, J.R., Pilkington ,R., Tait, K. and Tattersall, C. (1990). Question Interpretation and An-
swering. In J. Breuker, (ed.), EUROHELP: Developing Intelligent Help Systems; EC. 

Hintikka, J. 'and Hintikka, M.B. (1982). Sherlock Holmes Confronts Modem Logic: Toward a 
Theory ofInformation-Seeking Through Questioning. In Barth and Martens (J 982). 

Hobbs, DJ. and Moore, DJ. (I994). Multimedia Open Learning Systems in a New UK Univer-
sity. Proceedings of interdisciplinary Workshop on Complex Learning in Computer En-
vironments, Finland. 

Hoepelman J, Machate J, Schnitzer R ( 1991 ) Intonational Focusing and Dialogue Games; in 
Journal of Semantics, vol. 8, part 3: 253-275. 

Jackson, P. (1990). Introduction to Expert Systems. Addison-Wesley. 
Jeremaes, P. (1987). SpecifYing the Interface Logic. In H.J. Bullinger and B. Shackel, (eds.), 

Human-Computer Interaction-INTERACT '87. Elsevier Science Publishers B.V. 
Levin, J.A. and Moore, J.A. (1977). Dialogue - Games: Metacommunication structure for natu-

rallanguage interaction. Cognitive Science, vol. 1, part 4: 395-420. 
Lorenz ,K. (1982). On the Criteria for the Choice of Rules of Dialogic Logic. In Barth and 

Martens (i982). 



162 David Moore & Dave Hobbes 

Lorenz, K. (1987). Meaning Postulates and rules of Argumentation: Remarks Concerning the 
Pragmatic Tie between Meaning (ofTerms) and truth (of Propositions). In F.H. van Eemeren, 
R Grootendorst, J.A. Blair and C.A. Willard, (eds.), Argumentation: Across the Lines of 
Discipline. F oris Publications. 

Mackenzie, J.D. (1979a). Question-Begging in Non-Cumulative Systems. Journal of Philo-
sophical Logic: 117-133. 

Mackenzie, J.D. (I 979b). How to Stop Talking to Tortoises. Notre Dame Journal of Formal 
Logic, vol. XX, no. 4: 705-717. 

Mackenzie, J.D. (1980). Why do we Number Theorems? Australasian Journal of Philoso-
phy, vol. 58: 135-149. 

Mackenzie, J.D. (1981). The Dialectics of Logic. Logique et Analyse, vol. 94: 159-177. 

Mackenzie, J.D. (1984). Begging the Question in Dialogue. Australasian Journal of Philoso-
phy, vol. 62: 174-181. 

Mackenzie, 1.0. (1985). No Logic before Friday. Synthese 63: 329-341. 

Mackenzie, 1. (1990). Four Dialogue Systems. Studia Logica, XLIX, 4: 567-583. 
Mann, W. (1988). Dialogue Games: Conventions of Human Interaction. Argumentation vol. II: 

511-532. 

Maybury, M. (1992). Communicative Acts for Explanation Generation. International Journal 
of Man-Machine Studies, 37: 135-172. 

Moore, OJ. (1993). Dialogue Game Theory for Intelligent Tutoring Systems. Unpublished 
PhD dissertation, Leeds Metropolitan University. 

Mushrush, J.L. (1990). Options in Learning: Instructor led and CBT. Journal of Library and 
Administration, vol. 12, no. 2: 47-56. 

National Curriculum Council. (1990a). Curriculum Guidance 7 - Environmental Education; 
NCC,York. 

National Curriculum Council. (1990b). Curriculum Guidance 8 - Educationfor Citizenship; 
NCC, York. 

O'Shea ,T. and, Self, 1. (I983). Learning and Teaching with Computers: Artificial Intelli-
gence in Education. The Harvester Press. 

Parsons, M.W. (1985). Computers and Religious Education. In P.R. Smith, (ed.), Advances in 
Computer Assisted Learning, Proceedingsfrom the CAL 85 Symposium. Pergamon Press 

Pilkington, R.M. (1992a).Intelligent Help, Communicating with Knowledge Based Systems. 
Paul Chapman Publishing Ltd. 

Pilkington, R.M. (1992b). Question-Answering for Intelligent On-Line Help: The Process of 
Intelligent Responding. Cognitive Science 16: 455-489. 

Pilkington, R.M., Hartley, 1.R., Hintze, D. and Moore, OJ. (1992). Learning to Argue and 
Arguing to Learn: An Interface for Computer-based Dialogue Games. Journal of ArtifiCial 
Intelligence in Education, vol. 3 no. 3: 275-295. 

Reichman, R. (1985). Getting computers to talk /ike you and me: Discourse Context, Fo-
cus, and Semantics (An ATN Model). MIT Press. 

Ridgway, 1. (1988). Of Course ICAI is impossible-.... worse, though, it may be seditious. Self 
(1988). 



Computational Uses of Philosophical Dialogue Theories 163 

Self, J. (Ed.). (1988). Artificial Intelligence and Human Learning: Intelligent Computer-
Aided Instruction. Chapman and Hall. 

Shiffrin, D. (1985). Lauri Carlson: Dialogue Games, An Approach to Discourse Analysis (review 
article). Language in Society, vol. 144: 98-100. 

Shneiderman, B. (1992). Designing the User Interface - Strategies for Effective Human-
Computer Interaction. Addison-Wesley. 

Stenton, S.P. (1988). Dialogue Management for Co-operative Knowledge Based Systems. The 
Knowledge Engineering Review: 99-122. 

Stewart-Zerba, L. and Girle, R. (1993). Rules and Strategies in Dialogue Logic. Proceedings of 
the Sixth Australian Joint Conference on Artificial Intelligence, Melbourne. 

Tait, K. (1987). The Study Station Concept. University Computing, vol. 9: 25-28. 
Tattersall, C. (1993). Generating Help for Users of Applications Software. User Modelling and 

User-adapted Interaction: Special Issue on Natural Language Generation and Dialogue 
Planning. Kluwer. 

Walton, D.N. (1984). Logical Dialogue Games and Fallacies, University Press of America. 

Walton, D.N. (1985). New Directions in the Logic of Dialogue. Synthese, 63: 259-274. 
Walton, D.N. (1989). Question-Reply Argumentation. Greenwood Press. 
Woods, J. and Walton, D. (1982a). Argument: The Logic of the Fallacies. McGraw-Hill 

Ryerson. 
Woods, J. and Walton, D. (1982b). Question-Begging and Cumulativeness in Dialectical Games. 

Nous: 585-605. 

Yazdani, M. (1986). Intelligent Tutoring Systems Survey. Artificial Intelligence Review I: 43-52. 

D. J MOORE / D. J HOBBES 
FACULTY OF INFORMATION AND ENGINEERING SYSTEMS 

LEEDS METROPOLITAN UNIVERSITY, BECKETTE PARK 
LEEDS LS6 3QS 

UNITED KINGDOM