workshop on cognitive semantics and formal ontologies

Today I attended this workshop on formal ontologies and cognitive semantics organised in the context of the FOIS 2004: Formal Ontologies in Informatics Systems. There were almost 70 participants, but only 35 of them submitted a position paper. The contributions were organised in three panels around the following themes: Cognitive ontology engineering, Specific Approaches from Cognitive Semantics and Space and Time. Two brilliant keynote during the day: Formalisation and Implementation of Cognitive Semantics presented by Joseph Goguen and Transformations of Image Schemas, presented by Peter Gardenfors. The whole event was chaired by Werner Kuhn.

My running notes on the extended.

TITLE OF PAPER: FOIS Workshop on Cognitive Semantics in Ontologies
PRESENTED BY: misc, Werner Kuhn, Institute of Geoinformatics, University of Munster (Germany)

DATE: November, 3, 2004
LOCATION: Torino Incontra, Centro conferenze, Torino, Italy


Introduction (Nicola Guarino)
This workshop is very well aligned with the general conference theme.
Wener Kuhn
Why we are here -> reactions to the note he wrote during six months of workshop with Joseph Goguen.

Kuhn is not a computer scientist he is interested in how people term things in geographical information systems. He is an engineer and he is interested in solving problems for helping humans.
Some core questions: how meaningful are our models of semantics (e.g., in the Semantics Web)? How can cognitive semantics be used to make information systems more meaningful? In this process, time and place play a special role?

Some base line: information is for human beings: -information semantics needs to relate to meanings in minds – these meanings need observable effects (such as actions in the world resulting from understanding); there are different notions of meanings: -realist semantics; cognitive semantics, situated embodiments; “taking the best of both worlds” and integrating.

Dr. Joseph Goguen, professor of computer science at the university of california, san diego.
-formalization approach
-fuzzy logic
-formalization through logic

Keynote I: Formalization and Implementation of Cognitive Semantics (Joseph Goguen)

How to design human friendly ontologies? Mathematics, physics, formal philosophy are not always friendly!: -“top down” & often counter-intuitive; -culture specific -Barry Smith, John Sowa, Robert Kent. Most real ontologies built “bottom up”: B2b, ecology (EML), etc.

Why not using cognitively real constructs?: -basic level concepts; -basic image schemas; -combine with blending, etc.; -especially good for spatial ontologies. Evidence favors “middle-out” as humans: -Rosch.

Goals & methods of Cognitive Semantics: -understand language/mind/body interface; -understand concepts & meanings; – understand how mind works.
book: the way we think: conceptual blending …

Methods: -careful analysis od large bodies of language (spoken, written, graphics); -introspection (member’s competence).

George Lakoff

Rosh Experiments on Human Concepts
In is_a hierarchy, basic level is in meddle, has shortest name, most rapid, identification, most associated knowledge, earliest learned: -since has most human interaction
Highest level such that prototype exists, image representing whole category; similar motor actions for interaction with all instances.

Human concepts: examples: -shoe, not footwear, or sneakers; -apple, not fruit, or Machintosh.

Conceptual spaces, fremes & Domains: Fauconnier mental spaces are first order relational structures (mostly binary). Theories are better: declarations & axioms. Frame is densely interconnected systems of concepts (family father, mother, son, daughter. Domain is larger collection of more loosely connected concepts (e.g., law, education)

Lakoff Methaphor Theory: Image Schemas embodied and gestalt: -container; -journey; -in/out (is blend of two above)

——-X——> (image schemas)

Examples: he is trapped in his confusion
she can’t get out of her old habits

Metaphor is conceptual space map: -map is asymmetric: concrete source to abstract target; -map is partial: not all source used; -understand target via source; -both entities & inference mapped. Metaphor: “The sun is a king”. Metonymy: one thing stands for another e.g., part for a whole. Examples: “Paris disapproves our Iraqi policies.

Fauconnier & Turner: Conceptual Blending
-conceptual space networks
-simple blend diagram: input spaces & generic space
-some strong claims for blending: -is foundation of human thought; -including reasoning & perception; -is unconscious & rapid.
-but are many choices for blending: -so optimality principles are needed to decide among them.

-48 major blend for “house” and “boat”
-Oxymorons: -military intelligence (humor is re-blending)

Fauconnier & Turner: -use blending not mapping; -cross space map emergent from what’s in blend space

Promise & Problems: -experimental studies of gesture; -computer models os spatiality; -computational narratives

Poetry follows disoptimality principles.

Labov Structure in Extended BNF

Future Work: More on optimality principles; -semiotic blending for semiotic space (with levels & priorities); -Generalize architecture.

Panel A: Cognitive Ontology Engineering (Chair: Joseph Goguen)
<> Boyan Brodaric -> LEXIS (Logical conditions, schEma, content (K), conteXt, Individual, State.

<> Jorgen Fischer Nilsson -> Conceptual Spaces and Ontological Semantics Compositionally in Nominal Phrases

<> Dan Parvaz

<> Bill Pike -> The ontological bottleneck; the knowledge elicitation bottleneck; the tyranny of the majority. Codifying knowledge does not necessarily lead to successful application of that knowledge.
Hermeneutics – Semiotics – Situations
-How are the semantics of information resources revealed through action (manipulation, use, modification)?
-How do we build knowledge/context capture into the tools we already use?
-Is monitoring use cases a stap toward testability/falsifications?
-How do we describe why?
[this guy is very smart]

<> Joseph De Rosa (MITRE corporation) -> Cognitive Semantics, Complexity and Scale-free Networks. [a pedestrian view of something] Community sharing things in Space and Time: Can we dynamically “chunk” rich conceptual models, for simple, loose connectors, and create interoperability ad an emergent behavior?
[he try to bind conceptual spaces with complexity theory and Scale-free theory]

<> Csaba Veres -> Chomskian Linguistics, Semantics, and Information Systems. Jackendoff. He is trying to take the best of both worlds between Chomsky that is more on syntax and semantics. This Jackendoff (2002) was trying to report the fights between Chomsky and other linguists more interested in semantics.

Is there a conflict between people who are trying to do theoretical work and practical work? Why theorists are not using formal mathematics? This is something that we shouldn’t been fighting about.

What a semantic theory try to explain?

Chomsky -> syntax, automata, replication rules, …
What we are lacking here is a more engineering approach on how to handle the semantic data. [gardenfors]

There is a difference between an approach trying to look at the internal representation of a language, but for other the interest is more at the exterior usage of that language, on the outcomes.

The distinction between practice and theories is striking: statistics about the entailment of a particular representation.

We need some more engineering examples that force us to think practically and on the applicability of a certain theory. Can we show when and how a certain theory is better? Can we find good examples of this? Dogstulf site? [kuhn] {}

Some authors states that is not possible to formalise too much because concepts are too abstract. We can do this, we have to do that. What is happening in mind is much more of what we can put in a theory. [goguen]

We are lacking of an unified theory that bring all of this together. [aldo]

I don’t agree with aldo. The real problem is the transfer of the theory into an applied implementation of that theory.

A theory can be criticised only when it gets to the point of being formalised, so that you can see clearly what it cannot explain. [csaba]

Formalisation is not always a way to make a theory works. <-> we need more formalizations, principles, guidelines. To bridge the gap the theorists should work more on guidelines than on theories.

Formal ontologies are needed. We need primitives to give an account of the meaning. The theories are unavoidable. We need conceptual primitives and we need primitives extra beyond the conceptual ones. [guarino]

Keynote II: Transformations of Image Schema (Peter Gardenfors)
Prof. of Computer Science, Lund university.
Conceptual spaces, MIT Press.

Polisemy in language is a standard: many words have many meanings. Image schemas are usually represented as schematic pictures. It is an hard business to extract their meaning. Lanacker’s schemas. In a verb the focus in the time process. In a noun the focus in on the thing. For Lanacker moving from “to climb” to “climber” is just a transformation of the image schema.

What is an Image Schema?
The ontology if image schemas:
Domains: dimension in ‘conceptual spaces’
Sites: Trajectory, landmark, and other possible places for elaboration together with information about what categories they require.
Relations between sites: space and force dynamics.
Mark Jonhson -> force dynamics

Gardenfors thesis is that instead of calculating all the meaning of a certain schema, they can be extrapolated by transforming the image schemas.
Main operations on image schemas:
Elaboration: replacing an abstract site by a more specific schema.
Reprofiling: changing salience and focus of part of a schema. (Shift of attention)
-Metonymy: assigning trajectory status to parts or wholes of former trajectory
-Path: spatial vs. dynamic
-Multiplex: mass interchange (changing scale of resolution)
Metaphor: exchanging dimension/domain of the schema.

Two approaches to lexical meaning
Minimal specification: only central image schema is stored in the lexicon. Other meanings are obtained by transformations of this schemas.

If we get to specify more what is an image schema and the transformation that can be applied to these, then we would have amore systematic tool for working with semantics tools.

We can use logics to process the modifications. Logics is not enough for these transformations.

Can we find central meanings for image schemas? Like for the meaning of over. It is an empirical question.

I would use morphisms of Logics to translate your schemas transformation. Do we really need a logic. [gardenfors – goguen]

Panel B: Specific Approaches from Cognitive Semantics
<> Francesco Bellomi -> materialization(c,a) relates a more concrete individual c to a more abstract individual a which partially describes it. Formalization of materialization [Pirotte 1994, Dauchur 2004]: propose a formalizations of materialization which take into account only the inheritance of properties with concrete domain values. Their approach is to use a prototypes which provide a fictional instance of the object.

<> Giovanni Pezzullo, Aldo Gangemi -> Analysing Plans through Image-schemas and Descriptions: a multi-agent cognitive architecture with embedded constructive ontologies for managing …
Descriptions are assumed to represent the content of the epistemic use of a Cognitive Agent, and the process of applying a description …
They built a namespace called AKIRA
[I am lost]

<> Fabian Neuhaus -> Traditional position: the meaning of a unary predicate F can be made explicit by necessary and sufficient conditions (properties). Wittgenstein challenge is to say that something is true if there is a resemblance of some familiar tracts instead of the presence of properties.

The aim is the formalization of family resemblance.

<> Ian Stalker -> Dynamic Ontologies for Evolving Domains and Multiple Context
His position is not ti use fixed, singular definitions. Tokens convey information. Objects (individuals) as carriers. He wants to formulate a general lattice of meanings. The interesting things is that two conceptualizations map to two different specifications but then moving through this matrix we can map a concept with another concept.

<> Joost Zwarts -> ‘cognitive’ concepts: prototypical members, stereotypical attributes,
A formal context is a set of objects and a set of attributes and a relation between objects and attributes. -> a concept lattice. Representing prototype structure. Prototypical objects versus non-prototypical objects.


One one hand we have the web semantics, on the other we have the cognitive semantics with the cognitive schemas and all these sorts of things. Can we bring those together? Is there another foundations of the basics questions that leeds towards a more merged approach?

Where do we put inference mechanisms?

We need to choose the right primitives. We need a neutral language without formalism. Can we separate the inference mechanisms? [guarino]

Can we have new symbols attached to Image Schemata?

It is difficult to universalise something like “existence”.

Panel C: Time and Space (chair: Martin Raubal)
<> Tijana Asic (university of Geneva, CNRS Lyon)-> Blending for spatial and temporal representations. There must be something special for this spatial metaphor. Fauconnier call this middle spaces. The generic space is supposed to contain low level conceptual structures: what is common between space and time. Generic space is a simplification of the three dimensional space. In our thinking and talking we imagine time as mono-dimensional thing. In many other metaphors there in so such resemblance between space and time. Space and Time have lots in common, ontologically. What we need to think more seriously about blending of Fauconnier theory and a formal ontology.

<> Thomas Bittner (INFOMIS, Saarland University) -> Directly Depicting Ontologies. Simple expressions referring to parts, moments, and processes. very simple structures, very simple reasoning. DDO of endurants and perdurants -> levels of ontological theory.

<> Jenny Herding & Clare Davies, Ordnance Survey UK -> Concepts, semantics, and geographies. User needs, geographical concepts and semantics. What concepts are important to the user? Can models of human concepts and categories help? How well do categories apply to things in geographical space and time? Can basic level categories be identified?
– pedestrian wayfinding
– identifying areas of deprivation
Key dimensions in user conceptual views: entity concepts, identity name/function, extent, material, sensory quality, affordances, relationships, temporal qualities. Common dimensions in conceptual views for geographical space and time. Semantics for concepts and dimensions are context sensitive. Relative importance of dimensions in context dependent. Cognitive semantics in ontologies can improve the relevance to users.

<> James Hood (university of exter) -> Anchor Theory is a conceptual framework for the representation and reasoning about geographic objects with indeterminate boundary. This theory came about from a desire to: 1- overcome the limitation of current GIS that ties spatial objects to precise boundaries; 2- explore a solution to the problem of vagueness that does not depend …
Spatial objects whose boundaries is either unknown or indeterminate are anchored to spatial objects whose boundary …. Anchor theory Relationships: the hill is anchored in the hull summit, while the health Fritillary population is ANCHORED over the national park. Other forms of anchoring may include: viewpoint anchoring and temporal anchoring. What is we allow the ‘precise space’ to include objects whose boundary is only partially (precisely) defined? []

<> Werner Kuhn -> Image Schemas and Geospatial Ontologies: capture spatial meaning, provide grounding for abstract symbols, offer meaningful top level notions, primitive but not atomic, relate processes to entities, can be combined and transformed in powerful ways, are invariants in conceptual mapping.
A description of a word use some functions like: is contained. Different definitions can inherits features between each other. Ontologies need grounding: imager schemas offer it, at least for spatio-temporal task. Type classes can express combinations of image schemas: need no type casting.


Where do elements of spatial and temporal ontologie get their meaning from? How can we capture and represent cognitive and social context? Can geospatial and temporal concepts be represented mathematically?

Pro’s and com’s of cognitive + realist approaches to semantics.

I am not sure we need a meta-level of description of cognition in our ontology to describe space. We can partition space according to individual intent. [guarino]

Wrap-up session: where do we go from here? (Werner Kuhn)
Where do we go from here?
-as a book
-as a special issue of journal
-Suggested deadline for submissions: March
-Open call?

Leave a Reply