Reflections [VI] Conceptual Spaces At Work: Adams & Raubal - CSML

The final day of the CS@W conference started with a topic I was really anticipating as per a previous blog post. Ben Adams and Martin Raubal had indicated by email that CSML was a work in progress and a chat after their session emphasized the same. The existence of CSML and looking at some code examples holds great promise.

Direction in Lund, Sweden May 2012 by Peter Jones

The motivation for their work stems from geographical information systems [GIS] and how to formalise conceptual spaces. Palmer (1978) was cited as providing one form of representation. Palmer's mental model being matched with formal conceptual spaces. As per Gardenfors' book Minkowski's vector space is central to this work and informs a previous paper by Raubal (2004).

Way-finding provided an example: how there are different domains with semantic distances and weights that can change according to context. The facades of buildings and the contexts of day and night. What features stand out in terms of the shape, colour, visibility, area of buildings?

Adams and Raubal moved to differentiating topological relations that emphasise dynamic concepts. Piaget's contribution was noted in how we perceive conceptual change and the work of Torsten Hägerstrand whose time geography was illustrated (the geographical aspect of their work).

Behind CSML is a conceptual space algebra that is still being developed. On conceptual spaces and the semantic web, Adams and Raubal observed through their slides that:

Description logics are limited in ability to represent semantic content (RDF, OWL).
They fail to easily express
    Semantic similarity which plays an important role in cognitive categorization.
    Complex concept combinations.
Present Semantic Web technologies are restricted in their ability to answer queries and make inferences.

My BA(Philosophy and Computing Joint Hons.) project was on the application of semantic networks in nursing. The hard work was done for me in that it was directed by existing work in GRAIL. The dissertation I produced was essentially descriptive looking at nursing examples, identifying care issues and modelling them using is_a relationships. Among a great many slides and fascinating explanations Adams and Raubal note that:

Is-A, Instance-Of
Strong – based on topological relationships
Weak – based on similarity relationships

- so the advent of conceptual spaces, CSML and their efforts to engage with the existing semantic web community shows the ongoing dynamic of the semantic web's development. They are pursuing a reification of OWL 2 CS file to OWL 2:

Unique IRI for OWL classes, individuals, properties
Unique IRI for CS domains, concepts, and instances

There may be a future website to follow in addition to the CSML source. Questions followed about whole spaces, contrast classes and support vectors.

Reflecting back, my studies from 1996-97 and the examples produced - mobility... did not really even scratch the surface. Ben and Martin's work begs the question of trying to develop CSML examples that go beyond the surface(s) and enter the conceptual space (or spaces) that is Hodges' model.

Adams, B., Raubal, M. (2009). Conceptual Space Markup Language (CSML): Towards the Cognitive Semantic Web. ICSC '09 Proceedings of the 2009 IEEE International Conference on Semantic Computing IEEE Computer Society Washington, DC, USA.

(Photo: mine)