Section: New Results

Cultural knowledge evolution

Our cultural knowledge evolution work currently focusses on alignment evolution.

Agents may use ontology alignments to communicate when they represent knowledge with different ontologies: alignments help reclassifying objects from one ontology to the other. Such alignments may be provided by dedicated algorithms [4], but their accuracy is far from satisfying. Yet agents have to proceed. They can take advantage of their experience in order to evolve alignments: upon communication failure, they will adapt the alignments to avoid reproducing the same mistake.

We performed such repair experiments [3] and revealed that, by playing simple interaction games, agents can effectively repair random networks of ontologies or even create new alignments.

Modelling in dynamic epistemic logic

Participants : Manuel Atencia, Jérôme Euzenat, Line Van Den Berg [Correspondent] .

We explored how closely these operators resemble logical dynamics. We developed a variant of Dynamic Epistemic Logic to capture the dynamics of the cultural alignment repair game. The ontologies are modelled as knowledge and alignments as beliefs in a variant of plausibility-based dynamic epistemic logic. The dynamics of the game is achieved through (public) announcement of the game issue and the adaptation operators are defined through conservative upgrades, i.e. modalities that transform models by reordering world-plausibility. This allowed us to formally establish some limitations and redundancy of the operators [9]. More precisely, for a complete logical reasoner, the operators are redundant and some may be inconsistent with the agent knowledge.

These results hold for one agent in the game but not necessarily for the other that may not know the classes by which the alignment is repaired, nor the relations between them. The former can be dealt with by declaring that agents are aware of the signature of both ontologies (public signature assumption) but this does not allow ontologies to evolve. We are currently investigating partial semantics as a more dynamic alternative solution to this problem.

This work is part of the PhD thesis of Line van den Berg.

Populations

Participants : Manuel Atencia, Fatima Danash, Jérôme Euzenat [Correspondent] .

We started taking the population standpoint on experimental cultural evolution. For that purpose we introduced the concept of population within the experiments. So far, a population is characterised as a set of agents sharing the same ontology. Such agents play the same alignment repair games as before with agents of other populations.

The notion of population enables to experiment with different transmission mechanisms found in cultural evolution: vertical transmission, in which culture spreads, like genes, from parents to sibblings, and horizontal transmission, in which it spreads among all members of a population. We implemented explicit horizontal transmission through a synchronisation procedure in which, at a given interval, agents of the same population exchange their knowledge, i.e. alignments.

Link with interactor-replicator

Participant : Jérôme Euzenat [Correspondent] .

Cultural evolution may be studied at a `macro' level, inspired from population dynamics, or at a `micro' level, inspired from genetics. The replicator-interactor model generalises the genotype-phenotype distinction of genetic evolution. We considered how it can be applied to cultural knowledge evolution experiments [8]. More specifically, we consider knowledge as the replicator and the behaviour it induces as the interactor. We showed that this requires to address problems concerning transmission. We discussed the introduction of horizontal transmission within the replicator-interactor model and/or differential reproduction within cultural evolution experiments.

Experiment reproducibility

Participants : Jimmy Avae, Robin Couret, Jérôme Euzenat [Correspondent] .

Experiments are described and performed in our Lazy lavender platform which offers scripts to specify, run, and analyse experiments. This year, we investigated expressing experiment descriptions, i.e. design, results and analysis, in rdf . This facilitates the search of experiments based on structured queries that can be expressed in sparql : “which experiments have been performed but not analysed?”, “which experiments are derived from another specific experiment?”, “which hypotheses have not been confirmed since a precise release?”, “which experiments test F-measure increase?”. This also suggest a better organisation of our experiment reports.