derivations

[James Tauber]

HTMs have a time dimension, a spatial dimension and a hierarchical
dimension.

I suspect it's useful, then, to think about how different dimensions
of linguistic models map to those dimensions.

During the BOF, I talked a little about levels in a syntax tree and
their relationship to layers in an HTM, but someone (whose name I
didn't catch) pointed out the problem with that mapping is that syntax
trees can quite deep and you wouldn't want to rely on an HTM layer for
each level in the syntax tree.

But this morning it occurred to me that I hadn't considered the
*derivational* dimension. If one structure (say a syntax tree) derives
from another then perhaps it is those two structures that correspond
to two different layers in the HTM and that the entire syntax tree at
any given step in the derivation is entirely represented at one layer
in the HTM.

This also relates to our discussion at the BOF on the role of the
spatial dimension.

Of course that leads to the question of derivational versus constraint-
based approaches and which would be more applicable to HTMs.

It seems easier to imagine mapping a derivational approach to HTMs
than a constraint-based approached, and yet the latter have been more
successful in many areas of linguistics.

James

[James Tauber]

On Jun 27, 2008, at 8:03 AM, James Tauber wrote:
> HTMs have a time dimension, a spatial dimension and a hierarchical
> dimension.

FIgure 2.32 on p58 of Dileep's thesis illustrates this nicely.

James