Tony Veale & Mark T. Keane: Conceptual Scaffolding: A Spatially Founded Meaning Representation for Metaphor Comprehension. Computational Intelligence 8: 494-519 (1992)

@Article{,
  author = 	 {Tony Veale & Mark T. Keane},
  title = 	 {Conceptual Scaffolding: A Spatially Founded Meaning Representation for Metaphor Comprehension},
  journal = 	 {Computational Intelligence},
  year = 	 {1992},
  OPTvolume = 	 {8},
  OPTpages = 	 {494--519},
}

Author of the summary: Jim Davies, 2003, jim@jimdavies.org

Cite this paper for:

• Conceptual Scaffolding: "an interim meaning structure around which a fuller interpretation is fleshed out over time."
• System: TWIG knowledge-base management system
• neutral concepts [10]
• functional attributes[11]
• common attribute selection [31]

"Metaphors can be interpreted through core spatial metaphors forming a conceptual scaffolding between concepts. The role of the conceptual scaffolding is to create associations between ideas." [3]

Specifically, this paper looks at collocation, containment, and orientation.

System: TWIG knowledge-base management system

The broad semantic themes of an utterance are first constructed, then it's fleshed out, in two stages. [5] The fleshing out involves:

• label association with a conceptual relationship
• infer new associations
• establish coherence

• connect/disconnect: means association and dissociation ("Mary ran out of money" is a disconnect.) Symmetric.[6,7]
• actual causality/attempted causality: an agent node connects with one of these to a relationship, as a causal factor [8]. ("Bill gave Mary the book" = actual-cause(Bill,connect(Mary, book))
• contain/release: Specializations of connect/disconnect. Non-symmetric. ("Bill has fire in his veins." in is a containment cue. Same metaphoric meaning as "Bill is hot-blooded.") [9]
• up/down: Up(appearance) = beautiful, clean, etc.; up(speed) = fast. Where "appearance" and "speed" are neutral concepts that take on meaning with the up/down metaphor. [10]

More complexity: up(laptop) = down(laptop.weight) AND down(laptop.size) [12]

Laptop is a aggregation of attributes from laptop, computer, and product, and the context telss the agent which attributes to attend to.

[17] Another example:
"OS/2 lost a fortune"
disconnect(OS/2,Fortune)
disconnect(OS/2,up(wealth))
connect(OS/2,down(wealth))
down(OS/2)
down(OS/2.market-share)

It's hooked up to a KB, which takes as input a concept relation (e.g. connect(superman, clark-kent)) and outputs a e.g. connect clark-kent to superman as alter-ego.

attributes can be transferred with transfers, e.g. Transfer(fire->heat); Transfer(bath->clean). Where something with a connect between fire gets heat and connected with bath gets clean. [28]

Certain things will get transferred and others not depending on what those things are. e.g. "The laptop is Godzilla" means it's huge and slow, but not green and pungent. [30] Favored by common attribute selection and functional attributes of the target (called the tenor in this paper). [31]

I didn't get to the parts about coherence.

Differences with conceptual dependency theory:

• CD intends to capture all meaning, this just broad strokes.[38]
• because they are spatial metaphors, they are not as brittle as CD. (summary author's note: huh?)
• PTRANS, for example, is the same meaning for everything you put it to, where this bends meaning according to the associated words and concepts.
• CS is the starting point for elaboration, CD is the final interpretation.

Summary author's notes:

• Page numbers are from a preprint.