Language games for ontology alignment under partial observability

Merging information from different sources is crucial in the context of big data. For researchers, querying at the same time multiple heterogeneous databases would lead to many new insights by doing cross-database analysis or enhanced data integration, because different databases may contain diverse and complementary datasets. For large corporations, the alignment of the ontologies underlying the business processes of large companies in the case of mergers and acquisitions is a billion-dollar industry. While current methods for ontology alignment have been successful in identifying correspondences between concepts and relationships, it remains a very difficult task mostly due to the ”semantic gap” [1]. New research tries to tackle this issue by using knowledge graphs [2] but it remains an unsolved problem.

To overcome this difficulty, this project aims to use language as an abstraction layer over ontologies. More precisely, we will investigate how to align ontologies by using artificial agents, without the need for human intervention.

Our research involves conducting experiments with multiple agents, each being connected to a database. These agents engage in task-based communicative interactions [3], following the language game paradigm [4]. Identifying common information across their respective databases, they co-construct a shared language. This shared language maps for each individual database between the emergent language on the one hand, query language and the database on the other hand. The emerged language can then be used to query at once the different databases.

The ontology alignment we are talking about in this project operates at the language level. It uses artificial agents to share information across different databases. This project involves Fluid Construction Grammar (FCG) [5] and Incremental Recruitment Language (IRL) [6].

References :

[1] M. Ehrig, Ontology alignment: bridging the semantic gap, volume 4, Springer Science & Business Media, 2006.

[2] A. Hogan, E. Blomqvist, M. Cochez, C. d’Amato, G. d. Melo, C. Gutierrez, S. Kirrane, J. E. L. Gayo, R. Navigli, S. Neumaier, et al., Knowledge graphs, ACM Computing Surveys (CSUR) 54 (2021) 1–37.

[3] Nevens, J., Van Eecke, P., and Beuls, K. (2020). A practical guide to studying emergent communication through grounded language games. arXiv preprint arXiv:2004.09218.

[4] L. L. Steels, Evolutionary language games as a paradigm for integrated ai research, in: 2012 AAAI Spring Symposium Series, 2012.

[5] L. Steels, Introducing fluid construction grammar, Design Patterns in Fluid Construction Grammar. Amsterdam: John Benjamins (2011).

[6] M. Spranger, Incremental recruitment language — a formalism for evolutionary semantics, 2014, pp. 523–524. doi:10.1142/97898146036380125.