Meeting � Peter Gorniak, re Regier space model implementation

Greg Detre

@10.40 Thursday, October 10, 2002

 

Regier hasn�t considered extended trajectors

added problems with 3D

how do you represent it?

 

Regier assumes that there are 4 directions

prepositions shouldn�t be independent � e.g. you can have above-left etc.

 

C++ code/implementation

look at main � take out any reference to PVM

case of VISUAL-WORLD-STATE-MESSAGE-TYPE

ignore remaining loops � may not need main at all

PVM � ignore

very little PVM code inside the object

get rid of pack/unpack functions

make reginos by adding vectors of pixels my self

convex hull stuff � ignore

in the end, Peter said he�ll strip most of that stuff out

 

look especially at : region.h, .cc, pixel

 

imagine it�s a top-down view of the spatial world

next, in front of, moving

Mandarin vs horizontal temporal metaphors???

 

NNs

the advantage of a NN is that you can coopt its internal representations more easily though, right???

I think it might be really important to have the spatial-only model on moving (i.e. inherently/already temporally-extended) objects, so that it will be easier to add the time dimension later

maybe look at Grossman and trace rule stuff