knowledge. and make it possible for the agent to manipulate all elements stract ontology representation (AOR). This AOR then can be used to capture abstract models of communication reand multiple ontology representations. terconnected and heterogeneous. This suggests that future munication languages, multiple ways of expressing content of messages in a uniform way - as instances of its ontological agents will need to be able to deal with multiple agent comissues and describes a prototype implementation. an agent’s internal knowledge representation with an ablated knowledge (domain models, agent communication lanOne way to deal with this heterogeneity is by identifying Agent Environments are becoming increasingly open, inguages, content languages and models of how these interact) The paper outlines the approach, highlights interesting 1. INTRODUCTION 2. DEFINITIONS facts) and meta knowledge (knowledge about classes of entities). into instance knowledge (knowledge of individual Figure 1: An agent’s knowledge is usually divided similarities between these frameworks is beyond the scope are: DAML, OIL, UML, Frames (such as those currently the model chosen is in anyway \the best" for the combinaelements below. The following is the example AOR which will be used from now on: of this paper, a useful intersection appears to include the appeared to be necessary for a basic systems. Target ORs pressed in BNF/EBNF. Although a detailed study of the out the rest of the paper. The intention is not to argue that used in FIPA specications), simple language grammars exnes an abstract on tology representation which is used throughtion of target ORs but to draw out common aspects which The above are rather general statements, this section de3.1.2 Abstraction from Conceptual Models may be several useful abstractions and more than simply didate for use as the agent’s internal knowledge representaAbstract Ontology Representation (AOR). In general there sitions (encoding - conceptual model, conceptual models In the context of this paper it is referred to as an tion.2 unied conceptual model) to simply presen tation however. two abstraction steps. We focus on these two major tranThe product of the second step is then a potential can

3.1.1 Abstraction from Encodings 3.1.3 Example Abstract Ontology about languages it knows at the same level as domain knowlcompatible with the AOR dened in the previous section.

For a language based on an EBNF grammar a rst pass at generating the model could be done as follows: nipulate them. This enables the agent to treat knowledge An OR can be used to construct conceptual models of lanedge. The objective is to model languages in a formalism guages. A logical usage of this is to give the agent access gies can in fact be seen as abstract grammars for languages. to these language models at runtime and allow it to maAs Craneeld et. al. point out in [1] however, ontolo

Object Car

VW

Meta

Car

Vehicle

Truck Car

Vehicle SubClassOf SameClassAs 3.2 Abstract Agent Languages 3.2.1 Abstraction from Encodings

Figure 4 sketches how languages might be dened as three separate but linked ontologies. Although FOL cannot be 3.2.2 Abstraction from Conceptual Models 3.3 Building Multilingual Agents { A subset of FIPA-KIF [ 5 ], corresponding to SKIF (a limited form of KIF covering only KIF sentences). 4. IMPLEMENTATION 4.1 Overview 4.2 Examples of Operation 1. What should the AOR include/exclude?: As noted in dened. a profound eect on the agent system. Although it choices by agent toolkit developers will have a eect on how large numbers of agents may use of ontologies is \internal" and not shared with the outside world Section 3.1 the choice of internal representation has a lowest common denominator of understanding for a are internal representations and as such do not need given set of ontologies to ensure both can model the to be public. Agents working in mission critical areas others perception of the situation. 2. Do Agents need to share AORs?: In principle AORs however are likely to need to nd w ays of establishing 3. What types of Agent Languages can be represented?: scoping for example).

Since the AOR as described can in principle model functional and object oriented). How does this relate pects of languages which could be regarded as part of the conceptual model cannot be represented (variable the main features of a wide range of languages (logical, to other meta-modelling work (e.g. [1]) and what asconcepts?: Currently the AOR chosen allows class equivcepts in one ontology may be equivalent to a combina4. How can we cope with equivalences between groups of tion of concepts in another. alence, more generally however a combination of conhence this exibilit y does not appear very useful. For tance. lar for the subset of elements which correspond to FOL In fact the syntaxes of FIPA-SL and KIF are very simisyntaxes) this capability is likely to be of great imporother languages with other syntaxes however (or other 4.3 Resources 6. What happened to the semantics anyway?: As with \syntactic" relationships but say nothing about how support eectiv e semantic checking tools. beyond the immediate scope of the paper it would be semantics might be managed or enforced. Although interesting to see if the abstractions described would current language descriptions the conceptual models only capture the basic concepts of a language and their