TLAF – Three-Layer Argumentation Framework

Short Description Entities Deriving Supports and Attacks TLAF Ontology An Example Publications	Authors:
	Paulo Maio and Nuno Silva
	Affiliation:
	GECAD – Knowledge Engineering and Decision
	Support Group
	School of Engineering – Polytechnic of Porto
	Porto, Portugal
	Contacts:
	Tel: +351 22 834 05 24
	Fax: +351 22 834 05 24
	E-Mail: pam@isep.ipp.pt

Short Description

TLAF is an abstract argumentation framework that comprehends three modeling layers:

· Meta-Model Layer: it defines the argumentation entities and relations holding between them, maintaining domain independence;

· Model Layer: it defines the entities and their relations for a specific domain and negotiation community. I.e. it is the instantiation of the meta-model about the particularities of a domain of application, and according to a community’s perception of that domain;

· Instance Layer instantiates a particular model layer to capture the information passing between the participants in an argumentation process.

Figure 1. The three TLAF modeling layers

Entities

Argument

It is made up of three parts:  (i) a statement as conclusion, (ii) a set of statements as premises and (iii) a reasoning mechanism inferring the conclusion from premises.

On the Model Layer, when two arguments are related through the  relationship (say ) it means that argument affects (positively or negatively) argument . It also means that argument  might have as premise the statement concluded by argument. On the Instance-Pool Layer, depending on the arguments content this relation (i.e. ) is automatically specialized either in a support relationship () or in an attack relationship ().  

Arguments representing intentions must be specified as IntentionalArgument. During an argumentation process, participants must achieve consensus about intentions (e.g. which tasks must be executed where each task is an intention) independently of the reasons (i.e. beliefs) behind such intentions. Yet, the conclusion of an IntentionalArgument might be used (directly or indirectly) as premise of another IntentionalArgument.

Statement

It captures the meaning of domain data in the scope of the argumentation. This includes mandatorily the domain intentions, but also the desires and believes.

A statement can be concluded by different arguments, and serve as premise of several arguments.

Yet, a statement is in conflict with another statement when it implies or suggests that statement is not true. Since the conflict relation depends on the statements content, this relation is only established on the Instance-Pool Layer.

Reasoning Mechanism

It represents a reasoning mechanism such as a rule, a method, or a process that can be applied by an argument to draw a conclusion-statement from a set of premise-statements.

Deriving Supports and Attacks

TLAF automatically derives the support and attack relationships between argument instances (i.e. and  ) based on (i) the premises and conclusions of the argument instances, (ii) in existing conflict between statements and (iii) in the conceptual information of the model layer, namely the defined  relations.

Figure 2 and Figure 3 graphically depict the conditions under which a support or an attack relationship is established between two arguments respectively.

Figure 2. Conditions to establish a support relation between  two arguments

Figure 3. Conditions to establish an attack relation between  two arguments

As a result of applying these rules to the information previously depicted in Figure 1, two relations would be established:

· Argument  supports argument   (rule Figure 2a);

· Argument  attacks argument   (rule Figure 3a).

TLAF Ontology

To represent and exchange TLAF structures and TLAF instantiations, we suggest the use of ontologies. For that, we have developed an ontology in OWL that mimics the TLAF formalization, which is available here.

The TLAF Ontology corresponds to the Meta-Model layer in the sense that it defines the concept of Argument, Statement, Reasoning Mechanism and the relations between them.

To automatically derive the support and attack relation between arguments, the TLAF Ontology has been enhanced with some SWRL rules (see here).

The SWRL rules address both the intentional and the extensional parts of the ontology. While the TLAF Ontology was made to be extended, these SWRL rules capture the TLAF constraints in a way that it is not necessary (or advisable) to extend.

To properly exploit the TLAF Ontology and the defined rules, you need to install the developed SWRL Built-in (built-in definition and implementation).

Alternatively, an TLAF can be modeled in any object-oriented programming language (e.g. Java). In that sense, Figure 4 depicts a UML view of a possible TLAF implementation.

Figure 4. A UML view of a possible TLAF implementation

An Example

The following example illustrates the use of the TLAF on an abstract domain of application.

Model Layer Definition

Consider that according to the perception of a given community, our domain of application has three different types of arguments, each one applying a distinct reasoning mechanism and also concluding a distinct statement.

This scenario is graphically depicted in Figure 5 and modeled by means of an ontology here.

Figure 5. The Model Layer of the abstract domain of the example

Model Instantiation (Instance-Pool)

Consider that as a result of a particular argumentation interaction between two members (say  and ) of that community,  have instantiated the model as depicted in Figure 6 (available here as an ontology).

Figure 6. Partial Instance-Pool of the Model Layer previously defined

Deriving Arguments Relationships

According to the Instance-Pool, the following support and attack argument relationships are established (depicted in Figure 7):

·  ;

· .

Figure 7. Inferred support and attack relationships between argument-instances

You may check (for instance using Protégé 3.4.4) that these relations are automatically established by a rules engine (e.g. Jess) when it runs over the provided ontology.

Publications

· “A Three-Layer Argumentation Framework”, Paulo Maio and Nuno Silva, 1st International Workshop on Theory and Applications of Formal Argumentation (TAFA 2011) at IJCAI, July 2011, Barcelona, Spain.

This work is partially supported by the Portuguese MCT-FCT project COALESCE (PTDC/EIA/74417/2006).

This work is being published in conferences, workshops and other knowledge disseminating events.

Copyright remains with the authors.

This information should not be copy-pasted without acknowledging its origins.
Please contact the authors for information on how to use or reference this information.