formal argumentation and human reasoning

8/13/2019 Formal Argumentation and Human Reasoning

1/6

Towards Large Scale Argumentation Support on the Semantic Web

Iyad RahwanInstitute of Informatics

British University in DubaiP.O.Box 502216, Dubai, UAE(Fellow) School of Informatics

University of Edinburgh, UK

Fouad ZablithInstitute of Informatics

British University in DubaiP.O.Box 502216, Dubai, UAE

Chris ReedSchool of Computing

University of Dundee, UKDundee DD1 4HN, Scotland

Abstract

This paper lays theoretical and software foundations for aWorld Wide Argument Web (WWAW): a large-scale Web ofinter-connected arguments posted by individuals to expresstheir opinions in a structured manner. First, we extend therecently proposed Argument Interchange Format (AIF) to ex-press arguments with a structure based on Waltons theory ofargumentation schemes. Then, we describe an implementa-tion of this ontology using the RDF Schema language, anddemonstrate how our ontology enables the representation ofnetworks of arguments on the Semantic Web. Finally, wepresent a pilot Semantic Web-based system, ArgDF, throughwhich users can create arguments using different argumen-tation schemes and can query arguments using a SemanticWeb query language. Users can also attack or support partsof existing arguments, use existing parts of an argument inthe creation of new arguments, or create new argumentationschemes. As such, this initial public-domain tool is intendedto seed a variety of future applications for authoring, linking,navigating, searching, and evaluating arguments on the Web.

Introduction

A variety of opinions and arguments are presented every dayon the Web, in discussion forums, blogs, news sites, etc. Assuch, the Web acts as an enabler of large-scale argumen-tation, where different views are presented, challenged, andevaluated by contributors and readers. However, these meth-ods do not capture the explicit structure of argumentativeviewpoints. This makes the task of evaluating, comparingand identifying the relationships among arguments difficult.

Imagine querying the Web by asking List all argumentsthat support the War on Iraq on the basis of expert assess-ment that Iraq has Weapons of Mass Destruction (WMDs).You are presented with various arguments ordered based onstrength (calculated based on the number and quality of itssupporting and attacking arguments). One of these argu-

ments is a blog entry, with a semantic link to a CIA reportclaiming the presence of WMDs. You inspect the counterar-guments to the CIA reports and find an argument thatattacksthem by stating that CIA experts are biased. You inspectthis attacking argument and you find a link to a news articlediscussing various historical examples of the CIAs align-ment with government policies, and so on.

Copyright c 2007, Association for the Advancement of ArtificialIntelligence (www.aaai.org). All rights reserved.

Motivated by the above vision, we lay theoretical andsoftware foundations of a World Wide Argument Web(WWAW): a large-scale Web of inter-connected argumentsposted by individuals on the World Wide Web in a struc-tured manner. The theoretical foundation is an ontologyof arguments, extending the recently proposed ArgumentInterchange Format (Chesnevar et al. 2007), and captur-ing Waltons general theoretical account of argumentation

schemes (Walton 1996). For the software foundation, weimplement the ontology using the RDF Schema ontologylanguage (Brickley & Guha 2004) and present a pilot Se-mantic Web-based system, ArgDF, through which users cancreate arguments using different schemes and can query ar-guments using a Semantic Web query language. Users canalso attack or support parts of existing arguments, or useexisting parts of an argument in the creation of new argu-ments. ArgDF also enables users to create new argumenta-tion schemes from the user interface. As such, ArgDF is anopen platform not only for representing arguments, but alsofor building interlinked and dynamic argument networks onthe Semantic Web. This initial public-domain tool is in-tended to seed what it is hoped will become a rich suite of

sophisticated applications for authoring, linking, navigating,searching, and evaluating arguments on the Web.The paper advances the state of the art in computational

modelling of argumentation in three ways. First, it presentsthe first Semantic Web-based system for argument annota-tion, navigation and manipulation. Second, the paper pro-vides the first highly-scalable yet highly-structured argu-ment representation capability on the Web. Finally, the pa-per contributes to the recently proposed Argument Inter-change Format (AIF) ontology (Chesnevar et al. 2007) byextending it to capture Waltons argument schemes (Wal-ton 1996) and providing a complete implementation of theAIF in a Semantic Web language. If successful, the WWAWwill be the largest argumentation support system ever builtbecause its construction is not centralised, but distributed

across contributors and software developers in the model ofmany emerging Web 2.0 applications.

Related WorkArgumentation-based techniques have found a wide rangeof applications in artificial intelligence and computer sci-ence. An area that has witnessed significant growth isargumentation-support systems(ASS) (Kirschner, Shum, &Carr 2003). State-of-the-art ASSs suffer two main limita-


2/6

tions. Firstly, they usually support a small number of partic-ipants. Secondly, most of them target specific domains, suchas education (Rowe, Reed, & Katzav 2003) or jurisprudenceArguMed (Verheij 2003). Consequently, they are based onspecialised approaches to argumentation.

The World Wide Web can be seen as an ideal platform forsupporting large-scale argumentative expression and com-munication, due to its ubiquity and openness. On-line dis-cussion forums, such as Deme(Davieset al.2004), can pro-

vide a medium for such communication. However, whilediscussions may be identified by their topics, time, or par-ticipants, there is a lack of fine-grained structure that cap-tures the details of how different facts, opinions, and argu-ments contribute to the overall result. Having such structurecould enable better visualisation, navigation and analysis ofthe state of the debate by participants or automated tools.Automate support for argumentation may exploit the under-lying structure, for example, to discover inconsistencies orsynergies among arguments.

Recently, some Web-based tools have begun to enable

simple structuring of arguments. The truthmapping1 sys-tem supports a large number of participants but it only dis-tinguishes premises and conclusions, without providing a

distinction among different types of arguments, and withoutcross-referencing complex interactions among arguments. A

similar effort is being explored inDiscourse DB,2 which wasreleased to the public in late 2006. It provides a forum forcommentators to post their opinions about political eventsand issues. Opinions or arguments are organised by topic,and classified into three categories: for, against, and mixed.Moreover, content may be browsed by topic, author, or pub-lication type. Discourse DB is powered bySemantic Medi-aWiki(Volkelet al. 2006) and can export content into RDFformat for use by other Semantic Web applications.

Our aim in this paper is to combine the strengths of ASSsand the Semantic Web to enable highly-scalable yet highly-structured argument representation and processing capabil-ity in an open Web environment.

DesiderataWe propose a radically different approach to promotinglarge-scale argumentation. Instead of building yet anothersystem for supporting discourse among small or medium-size groups of participants, we aim to build an open, extensi-ble and reusable infrastructure for large-scale argument rep-resentation, manipulation, and (eventually) evaluation. Wenow list a set of key requirements that we believe are impor-tant to enable large-scale argument annotation on the Web.

1. The WWAW must support the storage, creation, updateand querying of argumentative structures;

2. The WWAW must have Web-accessible repositories;

3. The WWAW language must be based on open standards,enabling collaborative development of new tools;

4. The WWAW must employ a unified, extensible argumen-tation ontology;

5. The WWAW must support the representation, annotationand creation of arguments using a variety of schemes.

1Seehttp://www.truthmapping.com2Seehttp://discoursedb.org

Argument Interchange Format (AIF): Core

The AIF is a core ontology of argument-related concepts,and can be extended to capture a variety of argumentationformalisms and schemes. The AIF core ontology assumesthat argument entities can be represented as nodes in a di-rected graph called anargument network.

In the interest of simplicity, we shall use a set-theoreticapproach to describing the AIF. We will therefore use a set

to define each class (or type) of things like nodes. More-over, properties and relations between classes and instances(including graph edges) will be captured through predicatesover sets. Arguments are represented using a set N ofnodes connected by binary directed edges (henceforth re-ferred to as edges) which we define using the predicateedge

: N N. We will sometimes writen1edge

n2 to

denote(n1, n2) edge

. A node can also have a numberof internal attributes, denoting things such as textual details,certainty degree, acceptability status, etc.

The core AIF has two types of nodes: information nodes(orI-nodes) and scheme nodes (or S-nodes). These are repre-sented by two disjoint sets,NI N andNS N, respec-

tively. Information nodes are used to representpassiveinfor-mation contained in an argument, such as a claim, premise,data, etc. On the other hand, S-nodes capture the applicationofschemes(i.e. patterns of reasoning). Such schemes maybe domain-independent patterns of reasoning, which resem-ble rules of inference in deductive logics but broadened toinclude non-deductive inference. The schemes themselvesbelong to a class, S, and are classified into the types: rule ofinference scheme, conflict scheme, and preference scheme.We denote these using the disjoint sets SR,SC and SP, re-spectively. The predicate (uses:NS S) is used to expressthe fact that a particular scheme node uses (or instantiates)a particular scheme. The AIF thus provides an ontologyfor expressing schemes and instances of schemes, and con-strains the latter to the domain of the former via the function

uses. I.e., that n NS, s Ssuch that uses(n, s).The present ontology deals with three different types of

scheme nodes, namely rule of inference application nodes(or RA-nodes), preference application nodes(or PA-nodes)andconflict application nodes(orCA-nodes). These are rep-

resented as three disjoint sets: NRAS NS,NP A

S NS,andNCAS NS, respectively. The word application oneach of these types was introduced in the AIF as a reminderthat these nodes function as instances, not classes, of possi-

bly generic inference rules. Intuitively,NRAS captures nodesthat represent (possibly non-deductive) rules of inference,

NCAS captures applications of criteria (declarative specifi-cations) defining conflict (e.g. among a proposition and its

negation, etc.), and NP AS are applications of (possibly ab-stract) criteria of preference among evaluated nodes.

The AIF core specification does not type its edges. In-stead, semantics for edges can be inferred when necessaryfrom the types of nodes they connect. The informal seman-tics of edges are listed in Table 1. One of the restrictionsimposed by the AIF is that no outgoing edge from an I-nodecan be directed directly to another I-node. This ensures thatthe type of any relationship between two pieces of informa-tion must be specified explicitly via an intermediate S-node.


3/6

toI-node toRA-node toPA-node toCA-node

from I-node

I-node data usedin applying an

inference

I-node data usedin applying a

preference

I-node data in con-flict with informa-

tion in node sup-ported by CA-node

from

RA-node

inferring a

conclusionin the formof a claim

inferring a

conclusion in theform of aninference

application

inferring a

conclusion in theform of apreference

application

inferring a conclu-

sion in the form ofa conflict definitionapplication

from

PA-

node

applying apreferenceover data in

I-node

applying apreference overinference

application in

RA-node

meta-preferences:applying a

preference over

preferenceapplication in

supportedPA-node

preference applica-tion in supportingPA-node in conflict

with preference

application in PA-node supported by

CA-node

from

CA-node

applying

conflictdefinition todata in

I-node

applying conflict

definition toinferenceapplication in

RA-node

applying conflict

definition topreferenceapplication in

PA-node

showing a conflict

holds between aconflict definitiona nd some other

piece of informa-tion

Table 1: Informal semantics of untyped edges in core AIF

Definition 1 (Argument Network)Anargument networkis a graph consisting of:

a setNof vertices (ornodes); and

a binary relation edge:N N representing edges.

such that(i, j)edge

where bothi NI andj NI

A simple argument can be represented by linking a set ofpremises to a conclusion via a particular scheme. Formally:

Definition 2 (Simple Argument)Asimple argumentin networkis a tuple P,,c where:

P NIis a set of nodes denoting premises; NRAS is a rule of inference application node; and c NIis a node denoting the conclusion;

such that edgec, uses(, s) wheres S, and p P we

havep

edge

.Following is a description of a simple argument in propo-sitional logic, depicted graphically in Figure 1(a). We dis-tinguish S-nodes from I-nodes graphically by drawing theformer with a slightly thicker border.

Example 1 (Simple Argument)The tuple A1 = {p, p q},MP1, q is a simple ar-gument in propositional languageL, wherep NI and(p q) NIare nodes representing premises, andq NIis a node representing the conclusion. In between them,the node MP1 NRAS is a rule of inference applicationnodes (i.e., RA-node) that uses the modus ponens naturaldeduction scheme, which can be formally written as follows:

uses(MP1, A,B L A AB

B ).

An attack or conflict from one information or scheme nodeto another information or scheme node is captured through aCA-node, which captures the type of conflict. The attackeris linked to the CA-node, and the CA-node is subsequentlylinked to the attacked node. Note that since edges are di-rected, each CA-node captures attack in one direction. Sym-metric attack would require two CA-nodes, one in each di-rection. The following example describes a conflict, showngraphically in Figure 1(b), between two simple arguments.

pq

p

qMP1

(a) Simple argument (b) Attack among two simple arguments

r p

r

p MP2

neg1

A1

A2

pq

p

qMP1

neg2

Figure 1: Examples of simple arguments

Example 2 (Conflict among Simple Arguments)Recall the simple argumentA1 = {p, p q},MP1, q.And consider another simple argumentA2 = {r, r p},MP2,p. ArgumentA2 underminesA1 by support-ing the negation of the latters premise. This (symmetric)propositional conflict is captured through two CA-nodes la-belledneg1 andneg2.

Extending the Core AIF: Argument Schemes

Argumentation schemes are forms of argument, represent-ing stereotypical ways of drawing inferences from particularpatterns of premises and conclusions. Schemes help cate-gorise the way arguments are built. Among others, Waltonstaxonomy (Walton 1996) has has been most influential incomputational work. Each Walton schemetype has a name,conclusion, set of premises and a set of critical questionsbound to this scheme. A common example of Walton-styleschemes is theArgument from Expert Opinion:

Premise: SourceEis an expert in the subject domain S.

Premise: Easserts thatA, in domainS, is true.

Conclusion: A may plausibly be taken to be true.Many other schemes were presented by Walton (Walton1996), such as argument from consequence, and argumentfrom analogy. One can then identify instancesthat instanti-ate the scheme, such as the following example argument:

Example 3 (Instance of Argument from Expert Opinion)

Premise: Allen is an expert in sport.

Premise:Allen says that Brazil has the best football team.

Conclusion: Brazil has the best football team.

Critical questionsserve to inspect arguments based on thisscheme. For example, in the canonical scheme for Argu-ment from expert opinion, there are six critical questions:

1. Expertise Question:How credible is expertE?

2. Field Question: IsEan expert in the field that the asser-tion, A, is in?

3. Opinion Question:DoesEs testimony implyA?

4. Trustworthiness Question:Is Ereliable?

5. Consistency Question:Is Aconsistent with the testimonyof other experts?

6. Backup Evidence Question:Is Asupported by evidence?


4/6

As discussed by Gordon & Walton in the Carneades model(Gordon & Walton 2006), these questions are not all alike.The first, second, third and sixth questions refer topresump-tionsrequired for the inference to go through (e.g., the criti-cal question How credible is expertEas an expert source?questions a presumption by the proponent that ExpertE iscredible). The proponent of the argument retains thebur-den of proof if these questions are asked. Numbers fourand five, however, shift the burden of proof to the questioner

(e.g., the opponent must demonstrate that another expert dis-agrees with E). These questions capture exceptions to therule, and correspond to Toulminsrebuttal(Toulmin 1958).

Recall that in Example 1, we represented the rule of in-ference application in an RA-node labelledMP1, and statedexplicitly that it uses the modus ponens generic natural de-duction rule. It would therefore seem natural to use thesame approach with presumptive schemes. However, thisapproach loses the information about the generic structureof the scheme, as well as the explicit relationship betweenan actual premise and the generic form (or descriptor) itinstantiates (e.g. that premise Allen is an expert in sportinstantiates the generic form Source E is an expert in thesubject domainS). To this end, we propose capturing the

structure of the scheme explicitly in the argument network.We consider the set of schemes Sas nodes in the argu-

ment network. And we introduce a new class of nodes,called forms (or F-nodes), captured in the set NF N.Two distinct types of forms are presented: premise descrip-torsand conclusion descriptors, denoted byNPremF NFandNConcF NF, respectively. As can be seen in Figure 2,we can now explicitly link each node in the actual argument(the four unshaded nodes at the bottom right) to the form

node it instantiates (the four shaded nodes at the top right).3

Notice that here, we expressed the predicate uses with the

edge fulfilsScheme:NS S.

Since each critical question corresponds either to a pre-sumption or an exception, we provide explicit descriptions

of the presumptions and exceptions associated with eachscheme. To express the schemes presumptions, we add anew type of F-node called presumption, represented by thesetNPresF NF, and linked to the scheme via a new edge

type hasPresumption:S NPresF . This is shown in the three

(shaded) presumption nodes at the bottom left of Figure 2.As for representing exceptions, the AIF offers a more ex-pressive possibility. In just the same way that stereotypicalpatterns of the passage of deductive, inductive and presump-tive inference can be captured as rule of inference schemes,so too can the stereotypical ways of characterising conflictbe captured as conflict schemes. Conflict, like inference,has some patterns that are reminiscent of deduction in theirabsolutism (such as the conflict between a proposition and

its complement), as well as others that are reminiscent ofnon-deductive inference in their heuristic nature (such as theconflict between two courses of action with incompatible re-source allocations). By providing a way to attack an argu-

3To improve readability, we will start using typed edges, whichwill enable us to explicitly distinguish between the different typesof connections between nodes. All typed edges will take the formtype, wheretypeis the type of edge, and

type

edge.

mentation scheme, exceptions can most accurately be pre-sented as conflict scheme descriptions (as shown in the topleft of Figure 2).

Note that now, there is no longer any need to representcritical questions directly in the network, since they are eas-ily inferable from the presumptions and exceptions, viz., forevery presumption or exception x, that scheme can be saidto have a critical question Is it the case thatx?

Finally, in Waltons account of schemes, some presump-

tions may be implicitly or explicitly entailedby a premise.For example, the premise Source E is an expert in subjectdomain D entails the presumption that E is an expert in thefield that A is in. While the truth of a premise may be ques-tioned directly, questioning associated with the underlyingpresumptions can be more specific, capturing the nuancesexpressed in Waltons characterisation. This relationship,between some premises and presumptions, can be captured

explicitly using a predicate (entails:NPremF N

PresF ).

Definition 3 (Presumptive Inference Scheme Description)Apresumptive inference scheme descriptionis a tuple

PD, , cd,,, entails where:

PD NPremF

is a set ofpremise descriptors;

SR is the scheme; cd NConcF is a conclusion descriptor.

NPresF is a set ofpresumption descriptors;

SC is a set ofexceptions; and

entails NPremF N

PresF

such that:

hasConcDesccd;

pdPDwe have hasPremiseDescpd;

we have hasPresumption;

we have hasException;

With the description of the scheme in place, we can nowshow how argument structures can be linked to schemestructures. In particular, we define a presumptive argument,which is an extension of the definition of a simple argument.

Definition 4 (Presumptive Argument)A presumptive argument based on presumptive inference

scheme description PD, , cd,,, entails is a tuple

P,,c where:

P NIis a set of nodes denoting premises; NRAS is a rule of inference application node; and c NIis a node denoting the conclusion;

such that:

edgec; uses(, );

p Pwe havep edge;

fulfilsScheme;

c fulfilsConclusionDesc cd; and

fulfilsPremiseDesc P PD corresponds to a bijection(i.e. one-to-one correspondence) fromP toPD.


5/6

Conclusion descriptor:

A may plausibly be

taken to be true

Presumptive inference scheme:

Argument from expert opinion

Premise descriptor:

E is an expert in

domain D

Premise descriptor:

E asserts that A is

known to be true

Presumption:

E is credible as

an expert source

Presumption:

Es testimony

does imply A

Presumption:

E is an expert in the

field that A is in

hasPresumption entails

hasConclusionDescription

hasPremiseDesc

Conflict scheme:

Conflict from testimonial

inconsistency

Premise descriptor:

Other experts disagree

Conflict scheme:

Conflict from bias

Premise descriptor:

Speaker is biased

hasPremiseDescription

hasPremiseDescriptionhasException

hasException

Allen says that

Brazil has the

best football team

Allen is an

expert in sports

RA-node

Brazil has the best

football team

supportssupports

CA-node

CA_Node_attacks

Allen is biased attacks

fulfilsPremis

eDesc fulfilsPremiseDesc fulfilsPremiseDesc

fulfilsSchemefulfilsConclusionDesc

hasConclusion

Allen is not an

expert in sportCA-nodeattacks

I-node or a one of its sub-types

S-node or a one of its sub-types

F-node or a one of its sub-types

Scheme or a one of its sub-types

underminesPresumption

Underlined:Node type

Figure 2: An argument network showing an argument fromexpert opinion, two counter-arguments undermining a pre-sumption and an exception, and the descriptions of the

schemes used by the argument and attackers. A: Brazil hasthe best football team: Allen is a sports expert and he saysso;B: But Allen is biased, and he is not an expert in sports!

ArgDF: A Semantic Web System forAuthoring and Navigating Arguments

We implemented our extended ontology using RDF and

RDFS,4 and call the resulting ontology AIF-RDF. In sum-mary, we view elements of arguments and schemes (e.g.premises, conclusions) as RDF resources, and connect themusing binary predicates as described earlier.

ArgDF5 is a Semantic Web-based system built on the top

of the AIF-RDF ontology. It uses a variety of software com-ponents such as the Sesame RDF repository,6 PHP scripting,XSLT, the Apache Tomcat server,7 and MySQL database.

The system also uses Phesame,8 a PHP class containing a setof functions for communicating with Sesame through PHPpages. The Sesame RDF repository offers the central fea-

4The Resource Description Framework (RDF) is an XML-based language for making statements about resources. Each re-source has a Universal Resource Identifier (URI). A statement isa subject-predicate-object expression, sometimes called a triple.The predicate captures a relationship between the subject (a re-source) and the object (another resource, or a literal). RDF Schema(RDFS) (Brickley & Guha 2004) is a Semantic Web ontology lan-guage. It provides constructs for specifying classes and class hi-

erarchies, properties (or predicates) and property hierarchies, re-strictions on the domains and ranges of properties, etc. RDFSspecifications are themselves RDF statements. For example, thetriple (I-Node, rdfs:subClassOf, Node) specifies thatI-Node is a sub-class of Node.

5ArgDF is currently a proof-of-concept prototype and can beaccessed at: http://www.argdf.org

6See: http://www.openrdf.org7See: http://tomcat.apache.org/8

http://www.hjournal.org/phesame

tures needed by the system, namely: (i) uploading RDF andRDFS single statements or complete files; (ii) deleting RDFstatements; (iii) querying the repository using the SemanticWeb query language RQL; and (iv) returning RDF query re-sults in a variety of computer processable formats includingXML, HTML or RDF.

Creating New Arguments: The system presents theavailable schemes, and allows the user to choose the schemeto which the argument belongs. Details of the selectedscheme are then retrieved from the repository, and the formof the argument is displayed to the user, who then createsthe conclusion followed by the premises.

Support/Attack of Existing Expressions: The list ofexisting expressions (i.e. premises or conclusions) in therepository can be displayed. The user can choose an ex-pression to support or attack. When a user chooses to sup-port an existing premise through a new argument/scheme,this premise will be both a premise in one argument, and aconclusion in another. Thus, the system enables argumentchaining. If the user chooses toattackan expression, onthe other hand, s/he will be redirected to choose an appro-priate conflict scheme, and create a new argument whoseconclusion is linked to the existing conclusion via a conflict

application node (as in Example 2).Searching through Arguments: The system also en-

ables users to search existing arguments, by specifying textfound in the premises or the conclusion, the type of rela-tionship between these two (i.e. support or attack), and thescheme(s) used. For example, one can search for arguments,based on expert opinion,againstthe war on Iraq, and men-tioning weapons of mass destruction in their premises. Inthe background, the system construct an RQL query whichis then submitted to the RDF repository.

Linking Existing Premises to a New Argument: Whilecreating premises supporting a given conclusion through anew argument, the user can re-use existing premises fromthe system. This premise thus contributes to multiple ar-

guments in a divergent

structure. This functionality can beuseful, for example, in Web-based applications that allowusers to use existing Web content (e.g. a news article, a le-gal document) to support new or existing claims.

Attacking Arguments through Implicit Assumptions:With our account of presumptions and exceptions, it be-comes possible to construct an automatic mechanism forpresuming. ArgDF allows the user to inspect an existingargument, allowing the exploration of the hidden assump-tions (i.e. presumptions and exceptions) by which its infer-ence is warranted. This leads the way for possible implicitattacks on the argument through pointing out an exception,or through undermining one of its presumptions (as shownin Figure 2). This is exactly the role that Walton envisagedfor his critical questions (Walton 1996). Thus, ArgDF ex-

ploits knowledge about implicit assumptions in order to en-able richer interaction between the user and the arguments.

Creation of New Schemes: The user can create newschemes through the interface of ArgDF without having to

modify the ontology.9 This feature enables a variety of user-created schemes to be incorporated, thus offering flexibilitynot found in any other argument-support system.

9Recall that actual schemes are instances of the Scheme class.


6/6

Conclusions and Future Possibilities

As tools for electronic argumentation grow in sophistication,number and popularity, so the role for the AIF and its imple-mentations will become more important. What this paperhas done is to sketch where this trend takes us the WorldWide Argument Web and to describe some of the techni-cal components that will support it, building on a foundationof Waltons theory, the AIF and the Semantic Web. Earlier

in the paper, we introduced desiderata necessary for the cre-ation of a WWAW and we conclude here by revisiting them.

1. The WWAW must support the storage, creation, updateand querying of argumentative structures: ArgDF is aWeb-based system that supports the storage, creation, up-date and querying of argument data structures based onWaltons argument schemes. Though the prototype im-plementation employs a centralised server, the model cansupport large-scale distribution.

2. The WWAW must have Web-accessible repositories: Ar-guments are uploaded on an RDF repository which canbe accessed and queried openly through the Web and us-ing a variety of standard RDF query languages.

3. The WWAW language must be based on open standards,

enabling collaborative development of new tools: Argu-ments in ArgDF are annotated in RDF using RDFS on-tologies, both open standards endorsed by the W3C.

4. The WWAW must employ a unified, extensible argumenta-tion ontology: Our ontology captures the main conceptsin the Argument Interchange Format ontology (Chesnevaret al. 2007), which is the most current general ontologyfor describing arguments and argument networks.

5. The WWAW must support the representation, annotationand creation of arguments using a variety of argumenta-tion schemes: AIF-RDF preserves the AIFs strong em-phasis on scheme-based reasoning patterns, conflict pat-terns and preference patterns, and is designed specificallyto accommodate extended and modified scheme sets.

AIF represents a first step towards an open mechanism forrepresenting arguments, but the high level of abstraction thatwas demanded of it also presents challenges to developersabilities to use it. AIF-RDF bridges this gap between the on-tological abstraction and the code-level detail. ArgDF thendemonstrates the flexibility that AIF-RDF affords, and of-fers an example of rapid tool development on the basis oftheoretical advances in the understanding of argument struc-ture. Following are some potential usage scenarios that mayexploit the infrastructure presented here.

Question Answering:One extension of the current systemis to exploit the variety of Semantic Web techniques for im-proving question answering (McGuinness 2004). Prospectsrange from using query refinement techniques to interac-tively assist users find arguments of interest through Web-based forms, to processing natural language questions togenerate queries. This functionality would require annota-tions of a large amount content on the Web. Translating theontology to more expressive Semantic Web ontology lan-guages such as OWL (McGuinness & van Harmelen 2004)can also enable ontological reasoning over argument struc-tures, for example, to automatically classify arguments, orto identify semantic similarities among them.

Interface and argument visualisation: ArgDF providesonly rudimentary displays. More intuitive argument visu-alisation is needed for the WWAW to appeal to non-experts.

Argumentative Blogging: Another potential extension iscombining our framework with so-calledSemantic Bloggingtools (Cayzer 2004), to enable users to annotate their blogentries as argument structures for others to search, and toblog in response to one anothers arguments. This can po-tentially help build up large amounts of annotations, making

the question answering scenario above more viable.Mass-collaborative argument editing: Accumulating ar-

gument annotations can be done through mass-collaborativeediting of semantically connected argumentative documentsin the style ofSemantic Wikipedia(Volkelet al. 2006). Abasic feature of this kind is already offered by Discourse DB.

ReferencesBrickley, D., and Guha, R. V. 2004. RDF vocabulary de-scription language 1.0: RDF Schema. W3C Recommenda-tion REC-rdf-schema-20040210, W3C.

Cayzer, S. 2004. Semantic blogging and decentralizedknowledge management. Communications of the ACM47(12):4752.

Chesnevar, C. I.; McGinnis, J.; Modgil, S.; Rahwan, I.;Reed, C.; Simari, G.; South, M.; Vreeswijk, G.; and Will-mott, S. 2007. Towards an argument interchange format.The Knowledge Engineering Review21(4):293316.

Davies, T.; OConnor, B.; Cochran, A. A.; and Effrat, J. J.2004. An online environment for democratic deliberation:Motivations, principles, and design. Working paper, Sym-bolic Systems Program, Stanford University.

Gordon, T. F., and Walton, D. 2006. The Carneades argu-mentation framework. In Proc. 1st Intl. Conf. on Compu-tational Models of Argument, 195207. IOS Press.

Kirschner, P. A.; Shum, S. J. B.; and Carr, C. S., eds. 2003.Visualizing Argumentation: Software Tools for Collabora-

tive and Educational Sense-Making. London: Springer.McGuinness, D. L., and van Harmelen, F. 2004. OWLweb ontology language overview. W3C RecommendationREC-owl-features-20040210, W3C.

McGuinness, D. L. 2004. Question answering on the Se-mantic Web. IEEE Intelligent Systems19(1):8285.

Rowe, G. W. A.; Reed, C. A.; and Katzav, J. 2003. Arau-caria: Marking up argument. InEuropean Conference onComputing and Philosophy.

Toulmin, S. 1958.The Uses of Argument. Cambridge, UK:Cambridge University Press.

Verheij, B. 2003. Artificial argument assistants for defea-sible argumentation. Artificial Intelligence150(1-2):291

324.Volkel, M.; Krotzsch, M.; Vrandecic, D.; Haller, H.; andStuder, R. 2006. Semantic wikipedia. InProc. WWW 2006,Edinburgh, UK, 585594. ACM Press.

Walton, D. N. 1996. Argumentation Schemes for Presump-tive Reasoning. Mahwah NJ, USA: Erlbaum.

formal argumentation and human reasoning

Documents