Centre International de Recherche

sur l’Environnement et le Développement


Nos tutelles

CNRS Ecole des Ponts CIRAD EHESS AgroParisTech

Nos partenaires

R2DS MPDD FUTURS URBAINS LCS-R Net

Rechercher




Accueil > Rubrique de services > Archive Equipe > Minh Ha Duong > Opinions

IPCC risk and uncertainty guidelines seen from an Ontology Engineering viewpoint

publié le , mis à jour le

The efforts of the IPCC group on risk and uncertainty to build a common vocabulary [1] are closely related to a problem known in information technology marketspeak as Ontology Engineering. This suggests ways to improve the guidelines maturation process.

What Is Ontology Engineering ? [2]

As a branch of philosophy, Ontology is interested in the fundamental categories of being. But as it is used in computer science, an ontology is the product of an attempt to formulate an exhaustive and rigorous conceptual schema about a domain. This is roughly what the IPCC cross cutting groups are trying to do informally.

Ontology Engineering is a branch of knowledge engineering born with artificial intelligence and grown up from the need for a semantic web. That said, librarians did not wait for the invention of XML to start building thesauruses. Even long before that, philosophers already knew that the class of "humans" was a subclass of "animals".

An ontology defines a network of concepts. The most important relation between concepts is the subsumption link, also known as the specialisation/generalisation, subtype/supertype link, or put more simply the "is a kind of" relation... But other relations may be interesting, such as "part of".

When the network is a tree, we have a rigorously defined Taxonomy . Some people argue that any ontology should be taxonomic, but others favor a lattice structure. For example an ontology rigorously defines a Thesaurus structure [3] when it uses the "related to" link in addition to the "is a kind of" link. An ontology can sometimes be modularized as a set of trees.

An ontology is an explicit description of a domain :
- concepts
- properties and attributes of concepts
- constraints on properties and attributes

An ontology defines a common controlled vocabulary and a common conceptualization of the relationship between the concepts. It is used to share common understanding of the structure of information among people (and among software agents, that is integrated assessment models someday).

Making domain assumptions explicit makes it easier to understand and update legacy data, such as historical estimates of climate sensitivity and global warming.

An ontology also enables reuse of domain knowledge. This avoids “re-inventing the wheel” from a report to the next. Introducing standards allows interoperability, between IPCC working groups for example.

In "Ontology Engineering", the interesting part is not Ontology, but Engineering. Which social techniques to use to obtain an agreed upon vocabulary expliciting the implicit semantic networks used by people in a multinational ? How to formalize this vision of the world in its information infrastructure ? Leaving aside discussions about fundamental categories of being, the practical angle suggests that there are techniques to reach partial domain ontologies that work well enough. Which is what the IPCC risk and uncertainty coordination group wants.

Ontology Engineering and IPCC risk and uncertainty guidelines

Reading the litterature on OE [4] suggests a few ideas on how to improve the IPCC guidelines process :

- The IPCC Risk and Uncertainty guidelines are not going to be rewritten from scratch anytime soon. An ontology is a living object with a maintenance cycle : Evaluation -> Evolution -> Diffusion -> Use -> Evaluation.
- To improve the specifications of the ontology [5], use cases and scenarios could be systematically explored.
- The process would benefit from involving all parties, not only authors but also readers.
- It could be managed as a project : planified, documented, evaluated... There are no reason why the important work of building the ontology has to be done informally on the side by interested volunteer scientists.
- Data collection could go beyond brainstorming and discussion. Semi-structured interviews, document analysis and questionnaire can be used.
- Existing ontologies on the subject should be discovered, assessed and re-used as much as possible.
- Overall the IPCC Risk and Uncertainty ontology is informal. I think that it’s right to keep it informal since it’s mostly going to be used to write reports, not to feed computer agents. But the informal ontology should fit well with the mathematical theories of risk, uncertainty, decision, rationality... such as probabilities, imprecise probabilities, logic, game theory...

Conclusion

Maturing the IPCC Risk and Uncertainty guidelines is difficult because first, they have to make sense in a very large context : Physics, Engineering, Ecology, Political sciences... and second, the user community is loosely organised : thousand of scientists all over the world. On the other hand, it’s a small ontology of probably no more than 100 concepts, and it’s very much needed. Does this look like an international interdisciplinary research project ?


[1See Petit and Manning’s Scoping Paper, and this previous post.

[2Source : Ontology for the semantic web by N. F. Noy.

[3On the difference between thesaurus, taxonomy, ontology see this survey at metamodel.com.

[4Gandon (2002) Ontology Engineering : a Survey and a Return on Experience, especially chapters 2 and 3

[5Specifications for IPCC Risk and uncertainties have not been formalized and are rather fuzzy at this time. In Manning and Petit’s scoping paper, they can be found in the guideline’s first section. See also in Moss and Schneider’s guidance paper for TAR.