Meta-modelling for ontology development and knowledge exchange Mariano Fernández-López Laboratorio de Inteligencia Artificial Facultad de Informática Universidad Politécnica de Madrid Campus de Montegancedo sn. Boadilla del Monte, 28660. Madrid, Spain. Tel: (34-91) {336-66-05, 336-74-39} Fax: (34-91) 3-52-48-19 Email: mfernandez@fi.upm.es ABSTRACT Presently, methodologies do not propose to adapt One of the sources of heterogeneity of ontologies the mechanism of modelling to the different is that different ontologies have different ontologies to be built. However, our experience in necessities of modelling. This paper presents a bi- different projects (the (KA)2 initiative [Ben98], phase method to deal with these different the multidisciplinary project AM9819 about necessities. Phase I of the method models how to environmental pollutants, etc.) show that different model the ontology, obtaining a meta-model. Such domains should be modelled in different ways. meta-model can be expressed in LBIR, a formal Table 1 shows the components that have been and declarative language that has been specifically used in different ontologies. We can see that there designed for this task. To save resources, a are variations from some ontologies to others. reference meta-model that can be modified and Some ontologies have been built using a lot of reused is provided. During phase II of the method, attributes and no relations, others have been built the ontology is modelled following the meta- using constants, some of them have first order model obtained in the first phase. Furthermore, a logic formulas, but others do not, etc. tool (called ODE) provides software support to Apparently, one solution to this problem would be the method. Such tool generates SQL schemas to consider all the “necessary” components from LBIR, and allows the modelling of the (concepts, attributes, first order logic formulas, ontology following the selected meta-model. This constants, etc.) when an ontology had to be built. approach eases the interoperability between Nevertheless, such solution has the following groups located in different geographical locations drawbacks: (1) Our experience has shown it is that have to build the same ontology, since the possible that need for a component is not meta-model to be used can be exchanged through perceived a priori, that is, it is possible the LBIR. necessity of a component is only detected when it KEYWORDS is needed in an ontology. (2) New research about modelling can provide new components and new Ontology, meta-model, modelling, method, LBIR, ideas about how to use old components. (3) ODE. Considering non-useful components when an ontology is built can cause confusion in 1. EXPOSITION OF THE PROBLEM modellers, and especially when they are not very Even though Ontological Engineering is a very experienced. young area in Artificial Intelligence, there exist Besides flexibility in the components to be used some methodological proposals for building during the modelling, the knowledge should be ontologies: Uschold and King’s methodology presented in diffe rent ways to different experts. [Usc95], Grüninger and Fox’s methodology [Grü95], METHONTOLOGY [FeG99], etc. A Summarising, a rigid way to model brings us study and analysis of methodologies for building back to the classic knowledge-acquisition ontologies can be found at [Fer99]. This study bottleneck [Eri99]. shows that METHONTOLOGY is currently the most mature methodology. 1 Instance First order logic Arithmetic TOTAL Ontology Domain Concepts Relations Constants Instances attributes formulas formulas TERMS CHEMICALS.1 Chemical 10 6 0 0 0 1 20 37 CHEMICALS.2 Chemical 16 22 0 0 27 3 103 173 CHEMICALS.3 Chemical 16 20 0 2 27 1 103 169 Knowledge acquisition (KA) 2 restructured 78 12 47 0 0 0 102 239 community Reference Ontology Ontologies 23 70 9 0 0 0 8 110 Standard Units restructured Measure units 22 3 0 2 0 1 65 93 Monatomic ions Environmental ions 62 11 3 0 6 0 0 82 Silicates Silicates 84 17 8 0 0 0 0 109 Laboratory of Artificial Hardware 49 56 0 0 0 0 56 190 Intelligence's hardware ELLOS Ontology Catalogue of clothes 8 16 6 0 0 0 20 48 Catalogue of products in Tradezone Ontology 9 3 3 0 4 0 0 22 general SNCF Ontology Travels and hotels 13 37 4 4 2 2 1 69 FIDAL Ontology Contracts 6 15 8 0 1 0 7 37 Table 1. Components used in the ontologies developed with the bi-phase method 2 Figure 1. Concept classification tree in the domain of flights Table 2. Concept dictionary in the domain of flights concepts, attributes, first order logic formulas, Focussing on the case of METHONTOLOGY, it etc., and they are thought to be manipulated by proposes to carry out the following steps to experts in the domains to be modelled. Figure 1 develop an ontology: specification in natural presents an example of a graph: a concept language, conceptualisation using tables and classification tree, and table 2 is an graphs, formalisation (e.g. using frames), and implementation (e.g. using the Ontolingua example of concept dictionary. Tables language [Far97]). According to the and graphs in METHONTOLOGY are not fixed, METHONTOLOGY viewpoint, conceptualisation since the engineer can use tables or graphs that is the modelling at the knowledge level [New82], can be different to the proposed ones by the hence, the knowledge is modelled independently methodology. However, METHONTOLOGY of the implementation language to be used1 . The does not propose a precise way to specify how the proposed tables and graphs allow modelling tables and the graphs to be used during the conceptualisation are. Besides, this methodology 1 Such idea of conceptualisation is inspired in the Hayes-Roth and does not propose how to add a new type of table, colleagues’ approach [Hay83]. how to add a new field to a type of table, how to 3 delete one of the types of the proposed graphs, or In the following sections, a solution to these how to elaborate a completely new modelling way problems will be presented. Section 2.1 will with completely new graphs and tables. Therefore, present the bi-phase method and, section 2.2, its if several groups in different locations have to software support: ODE. The paper will finish with build an ontology collaboratively, there are the conclusions and future trends. problems to agree and exchange the characteristics of the tables and graphs to be used (see figure 2). Figure 2. Problems in collaborative construction when the characteristics of tables and graphs are not clearly specified conceptualisation, meta-formalisation and 2. THE PROPOSED SOLUTION meta-implementation. On the other hand, phase 2.1. THE METHODOLOGICAL LEVEL OF II carries out the specification, conceptualisation THE BI-PHASE SOLUTION (following the meta-model obtained in phase I), formalisation and implementation of the ontology. To allow a more flexible modelling of ontologies As you can see, in this bi-phase method, there is a and to ease the exchange of characteristics of modelling both at Newell’s knowledge level and tables and graphs, the bi-phase method proposes symbolic level during phase I as well as phase II. to model how to model the ontology. Until now, the purpose of the ontology engineer was to model To facilitate the building of meta-models, a some parts of the world, for example, flights, reference meta-model is proposed. It is possible chemical elements, etc. (see figure 3), however, to modify this reference meta-model according to with the bi-phase method, modelling the process the modelling needs of each ontology. Such meta- of modelling is also recommended, that is, model is expressed by means of meta-tables and building a meta-model is also proposed. meta-graphs, and it is also formally expressed. Particularly, the part of the modelling process to The reference meta-model allows building model is the conceptualisation, which is the base ontologies with: concepts, class and instance of the remainder steps of the modelling. attributes, facets of such attributes, relations, first order logic formulas, arithmetic formulas, The bi-phase method follows the constants, and instances. These components METHONTOLOGY approach, although in two appear in the reference meta-model because each levels. On the one hand, during phase I, the one of them have been used in some of the ontology conceptualisation process is specified in ontologies developed during the experimentation. natural language, conceptualised using tables and Besides, we have checked that the reference meta- graphs (called in this phase meta-tables and model contains the static components of the meta-graphs), formalised using a formal classic languages for ontology development language, and implemented in SQL (see figure 4). (Ontolingua, OKBC, OCML, FLogic and Thus, the result of this first phase is a meta-model LOOM). We say static components because we do presented in meta-tables and meta-graphs, in a not consider rules and procedures. This reminds as formal language, and in SQL. The steps of this future work. phase are called: meta-specification, meta- 4 Figure 3. General overview of the ontology development using meta-models We have also developed a tool, called ODE, in However, it is not the only tool allowing flexible order to provide software support to the bi-phase modelling, since Protégé-2000 [Fri00] permits the method. Ode is especially designed to facilitate user to redefine its components (made by classes, the application of the method. slots, etc.). Figure 4. Bi-phase method to build ontologies In [Fer01] a complete description of the method is presented. Such description includes the tasks to 5 be performed, the inputs, the outputs and the in an exhaustive partition3 . Besides, participants. This description includes a way to it can be also specified that a table to use during manage the changes in meta-models, even when the conceptualisation of the ontology is the an ontology is being developed with such meta- concept dictionary. The possible fields of model and new necessities are detected. There is such table would be: concept name, also a description of the architecture of ODE. The instances, instance attributes, etc. method and the tool have been tested in the above Concerning the recommended order, it should be mentioned projects (the (KA)2 initiative, the said that the elaboration of the concept multidisciplinary project AM9819 about classification tree should begin before starting the environmental pollutants, etc.). 10 different meta- concept dictionary. And with regard to the models have been built with a total of 33 consistency verification rules between the concept additions, removals and modifications with classification tree and the concept dictionary, all regards the reference meta-model; such meta- the concepts of the tree should be in the concept models have been used in 11 different domains: dictionary and vice versa. chemical elements (169 terms with 27 first order formulas), knowledge acquisition community (239 2.1.2. Meta-conceptualisation of the terms with no first order formulas), hardware (190 conceptualisation process terms with no first order formulas), ontologies (110 terms with no first order formulas), measure For (meta-)conceptualising in phase I, the bi- units (93 terms with no first order formulas), phase method proposes: (a) a set of meta-tables to monatomic ions (82 terms with 6 first order describe the tables and graphs to be used during formulas), silicates (109 terms with no first order the conceptualisation in phase II; (b) a meta-graph formulas), catalogues of cloths (48 terms with no to describe the order in the conceptualisation in first order formulas), travels (22 terms with no phase II; (c) and meta-tables and meta-graphs to first order formulas), hotels (69 terms with 2 first describe the consistency verification rules. Thus, order formulas) and contracts (37 terms with 1 for example, the meta-tables of node first order formula). Other meta-models have been description, and the meta-tables of built containing meta-graphs and meta-tables to edge description are proposed to define the model databases, other meta-models contain meta- details of the graphs, and the meta-tables of graphs to model tasks, and other meta-models field description are proposed to define even contain schemas of bills, invoices, etc. as the details of the tables. For instance, meta-tables meta-tables. 1, 2 and 3 show the description of the taxonomy and of the concept dictionary, used both in the In the following sub-sections, a brief description examples of section 1.1.1. In all these meta-tables, of the steps of phase I will be presented. the meta-field symbol is filled in with 2.1.1. Meta-specification of the abbreviations. In the case of meta-table 2, which conceptualisation process describes a graph, the meta-fields input and output edges, input multiplicities During phase I, the meta-specification describes, and output multiplicities are used to in natural language: (a) what tables and graphs establish how many edges can go in and go out to will be used during the conceptualisation of the and from a node. In the case of meta-table 3, ontology; (b) the recommended order to fill in the which describes the concept dictionary, the meta- tables and to build the graphs; and (c) the field format restricts the possibilities to fill in consistency verification rules between tables, the cells (text, list, logic expression, etc). Is it between graphs, and between tables and graphs. main is true when the described field is the For example, it can be (meta-)specified that a identifier of the row. Repetition in the graph to be used during the conceptualisation is same table is true when the field can be filled the concept classification tree, that in with the same value in different rows. And the nodes of this graph are concepts, and that multiplicity is true when the same cell can the edges are subclass of, subclass in have several values. a disjoint partition2 , and subclass 3 ‘Subclass in an exhaustive partition’. An exhaustive partition of a class is a set of subclasses that covers all the class, that is, there is not an 2 ‘Subclass in a disjoint partition’. A disjoint partition of a class is a set instance of the father class that is not an instance of any of the of subclasses of this class that do not have common instances. subclasses of the partition. 6 Edge Symbol Description Subclass of S A class C is a subclass of the parent class To carry out the meta-formalisation, a formal and P if and only if every instance of C is declarative language, called LBIR (Language for also an instance of P. Building Intermediate Representations), has been Subclass in a SDP A disjoint partition of a class is a set of disjoint its subclasses where the subclasses do elaborated. Such language has the same partition not have common instances. expressiveness as the meta-tables and meta-graphs Subclase in an SEP An exhaustive partition of a class is a set used during the meta-conceptualisation. The LBIR exhaustive of subclasses that covers all the class, partition that is, there is no instance of the father description uses a context free grammar for the subclass that is not subclass of any class syntax, and matrices to establish the meaning of of the subclasses of the partition the language. The following code: Meta-table 1. Meta-table of edge description defining the possible edges of the graph “concept classification tree” define table horizontal [Concept dictionary] as CD Node Symbol Descrip- Input Input Output define field [Concept name] as CN tion and multipli- multipli- begin outpud cities cities type term; edges repeated no ; Subclass (0, n) (0, n) multiplicity (1,1); of end field ; Concept C ** Subclass (0, n) (0, n) define field Instances as I in a disjoint begin partition type term; Subclass (0, n) (0, n) repeated yes; in an multiplicity (0,N); exhaus- define field [Instance attributes] as IA tive parti- begin tion type term; repeated yes; Meta-table 2. Meta-table of node multiplicity (0,N); description defining the possible nodes of the graph end field ; define field Relations as R “concept classification tree” begin type term; The meta-graph to model the order during the repeated yes; multiplicity (0,N); conceptualisation is not presented due to the space end field ; constraints. Concerning the consistency begin verification rules between tables, between graphs, placed in [Binary relation diagram]; and between tables and graphs, the way to write main field [Concept name]; end table ; them is based on operations on matrices representing the tables and the graphs. Such shows the definition in LBIR of the concept operations are similar to the ones used in the dictionary, that is equivalent to the definition relational model for databases (projection, appearing in meta-table 3. Placed in binary selection, difference, etc.). relation diagram indicates that a graph 2.1.3. Meta-formalisation and meta- called binary relation diagram should be designed before filling in he concept dictionary. implementation of the conceptualisation process Field Symbol Description Format Is it main? Repetition in the same table Multiplicity Concept name CN ** Term Yes No (1, 1) Instances are particular Instances I cases of the concept Term No Yes (0, n) The ones that allow Instance attributes IA describing the instances of Term No Yes (0, n) the concept. Relations R Relations link concepts Term No Yes (0, n) Meta-table 3. Meta-table of field description defining the table “concept dictionary” schema. This eases the use of databases to store During the meta-implementation, the meta-model ontologies, taking advantage of the independence expressed in LBIR is transformed into a SQL and integrity of the data, the minimisation of the 7 redundancy, etc., provided by the relational figure 6). The second option, LBIR, is mandatory database systems. if the current version of ODE is utilised to build the ontologies. 3.2. THE SOFTWARE LEVEL OF THE BI- PHASE SOLUTION The method and the tool have proved useful in several Spanish and international projects. In order to allow the efficient use of the methodology proposed in 3.1, we has built ODE (see figure 5). The LBIR processing module automates the transformation, without loss of expressiveness, from a meta-model in LBIR to a Meta-model in LBIR Meta-model in LBIR Meta-model in LBIR SQL schema. Besides, it allows conceptualising ontologies following a meta-model selected by the user, and storing the result in a database following LBIR PHASE I the SQL schema associated to the meta-model. processing Moreover, if you follow the reference meta-model to conceptualise your ontology you can use a SQL-schema generator of Ontolingua code. The main feature of SQL-schema SQL-schema ODE is that a change in the meta-model does not force a change in the program, since SQL schemas SQL are generated in run-time and not in design time schema Conceptualisation Ontolingua (as usual). copying process translator 3. CONCLUSIONS AND FUTURE TRENDS Knowledge to be conceptualisation Ontolingua code conceptualised result Database creation Although each ontology has its modelling needs, there is not any methodological proposal to use a conceptualisation content different kind of modelling for each ontology. PHASE II The bi-phase method presented in this paper Database to store the conceptualisation proposes, during a first phase, to model the modelling process itself (or reusing an existing meta-model) and, during the second phase, to model the ontology. In the first phase, the steps Figure 5. ODE processes are: meta-specification, meta-conceptualisation, One of the most interesting future lines, above all meta-formalisation and meta-implementation. for ODE, would be the fast development of During the second phase the steps are the ones translators from different meta-models into proposed by METHONTOLOGY: specification, different implementation languages. An interface conceptualisation, formalisation and to manipulate meta-tables and meta-grpahs would implementation. To carry out the meta- be also interesting. Another important future trend formalisation, a formal and declarative language would be a structured characterisation of (LBIR) has been elaborated. Moreover, to provide ontologies according to their modelling needs. software support to both phases, a tool has been Now, the modelling necessities are determined by developed: ODE. the experience of the ontology engineers, who To agree in the meta-model to be used, the interacts with the experts in the domain to be different groups can exchange this meta-model in modelled. meta-tables and meta-graphs, or in LBIR (see 8 Figure 6. Use of meta-models for cooperative construction of ontologies [FeG99] Fernández-López, M.; Gómez-Pérez, A.; 4. REFERENCES Pazos-Sierra, A.; Pazos-Sierra, J. [Ben98] Benjamins, V.R.; Fensel, D.; Gómez- “Building a Chemical Ontology Using Pérez, A.; Decker, S.; Erdmann, M.; METHONTOLOGY and the Ontology Motta, E.; Musen, M. 1998. “The Design Environment”. IEEE Intelligent Knowledge Annotation Initiative of the Systems & their applications. Knowledge Acquisition Community January/February. Pages 37-46. 1999. (KA)2 ”. In: Proceedings of the 11th Banff Knowledge Acquisition for Knowledge- [Fer99] Fernández López, M. “Overview of Based System Workshop (KAW 98), Methodologies for Building Ontologies”. Banff, Canada. Workshop on Ontologies and Problem- Solving Methods: Lessons Learned and [Eri99] Eriksson, H.; Fergerson, R.W.; Shahar, Future Trends. (IJCAI99). August. 1999. Y.; Musen, M.A. “Automatic generation of ontology editors”. Knowledge [Fri00] Fridman Noy, N.; Fergeson, R.W.; Acquisition Workshop (KAW). Banff Musen, M.A. “The knowledge model of (Canada). 1999. Protégé-2000: combining interoperability and flexibility”. Workshop on Knowledge [Far97] Farquhar, A.; Fikes, R.; Rice, J. “The Acquisition, Modelling and Management Ontolingua Server: Tool for (EKAW) . Editor Springer Verlag. Jean Collaborative Ontology Construction”. Les Pins (Francia ). 2000. IJHCS, 46(6) (1997) 707-728. [Grü95] Grüninger, M.; Fox, M. S. “Methodology [Fen00] Fensel, D. Ontologies: silver bullet for for the design and evaluation of Knowledge Management and Electronic ontologies”. Workshop on Basic Commerce. Springer-Verlag. Berlin. Ontological Issues in Knowledge 2000. Sharing. Montreal, Canada. 1995. [Fer01] Fernández-López, M. Método bi-fase [Hay83] Hayes-Roth, F.; Waterman, D. A.; Lenat, para la conceptualización de ontologías D. B. “Building Expert Systems”. basado en meta-modelos. Ph. Thesis. Addison Wesley Publishing Company, Facultad de Informática. Universidad Inc. Massachusets , USA. 1983. Politécnica de Madrid. Spain. 2001. [Nec91] Neches, N. ; Fikes, R. ; Finin, T. ; Gruber, T. ; Patil, R. ; Senator, T. ; 9 Swartout, W.R. “Enabling technology for knowledge sharing”. AI Magazine, Fall 1991. Pages 36-56. 1991. [New82] Newell, A. “The Knowledge Level”. Artificial Intelligence. Volume 18. Number 1. Pp. 87-127. January 1982. [Usc95] Uschold, M. King, M. “Towards a methodology for building ontologies”. Workshop on Basic Ontological Issues in Knowledge Sharing. International Joint Conference on Artificial Intelligence (IJCAI). Montreal, Canada. 1995. 10