Attended a thought-provoking seminar last week given by Robert St. Amant on
Physical Metaphors for Problem Solving.
seminar thoughts
Metaphors shape the way we interface with machines. they can be useful (in exposing intuitive interfaces), but also can be wrong/halfmapped. For example, on computers, we use “windows” as metaphors–but while some of our uses of GUI windows are in keeping with this metaphor (“open”, “close”), other actions are not (you don’t “look through” GUI windows. But you do “drop things” into them).
image schemas:
- mental pattern that provides an archetype for forms of interaction with the environment (metaphorical or real)
- they encapsulate specialized interrelational logic/representation
- a sizable subset of cognitive scientiests are pretty sure image schemas are the Language of Thought (e.g. Piaget)
- cognitive linguists, in particular, have developed Image Schema theory a lot. They say it does a very good job of explaining why people say what they do
- “most work in image schemas have been using them as post-hoc explanations for why people say things the way they do”. no work has been in using them to predict actions or to represent new knowledge.
- the major gains in using image schemas are
- generalization: chess problem-solving tactics apply to war problem-solving tactics, and vice versa
- robustness of representation: humans look at the world this way, it helps for consistency of interaction if we can get computers to look at the world this way too)
- This is where St. Amant’s work comes in. Cognitive linguists have made fine catalogs of different schemas, and he is using those schemas to create an actual language for knowledge representation purposes (much like Cyc has done for commonsense reasoning).
- Tangent:
- tangent question:how is this different from typical logical representation of reality /commonsense reasoning a la cyc?
- it looks like, as he creates these schemas, he’s just being explicit about using image schemas to codify–whereas with Cyc, they still create these using these schemas, but more subconsciously(?)
- Question to self: they are using chess and war-games a lot, as application. I can understand war games because of military funding… but why are they doing Chess? It seems like Go would be a much better application area for this!
- System evaluation work: not much done yet, as the project is still not very mature. There are a couple of metrics we could use to evaluate the system:
- how good is the expressive power?
- how good is the predictive power?
- how good does generalization work? this is what it comes down to. does this abstractified/metaphorical representation help with generalization for solving problems with this stuff?
- or, how closely does this representation fit with the way that humans represent knowledge? (can you use this abstract knowledge to instantiate situations in different fields (chess, war), and see if human experts are good at processing, or remember instances of this?)
- note that context defines schema. Looking at a room:
- a seminar attendee views a room as a location
- a fire marshall views a room as a container (one that can only hold X number of people)
- a painter views a room as a set of surfaces
- Note: schemas as they are now seem biased to represent nouns rather than verbs. The schema ontology reflects this. This could be an artifact of the things they are used to represent.
Related Wikipedia links: