Dick:
> Jasper replying to And:
>
> >I was suggesting that this simulated annealing has some intuitive
> >psychological plausibility, which conflicts with Dick's argument for a
> >parser that doesn't lead you up garden paths
> ## What I really believe about parsing is that it's done by spreading
> activation, so that the winning parse at any given moment is simply the
> most active; other alternatives have some above-rest activity, so it's not
> that much of a jump to increase their activity and make them win. You get
> garden-path pain if the (correct) alternative has returned almost to rest
> level, so you really have to go back to the beginning and start again. I
> wouldn't want to defend a model in which there are no garden paths,
because
> garden paths clearly exist. In some ways this view of parsing is like
> simulated annealing, in that it's heavily quantitative and probabilistic

Ah, this is very plausible.

> > > The supposed subject prototype is then merely a result the statistical
> > > prevalence of constructions in which the same word is True Subject,
> > > 'Object' and 'Agent-dependent'
> ## Precisely. There's a strong tendency for these properties (which you're
> assigning to distinct relations) to correlate. But concepts are how we
> express correlations among properties - e.g. we bring together wings,
eggs,
> nests, beaks, etc. by creating the concept Bird. So we notice and express
> this correlation by creating a super-relation called Subject. Whether we
> represent the properties as sub-relations as well I don't know; but we
must
> recognise the exceptional sub-cases where the properties part company and
a
> subject lacks one or more of these properties

Fair enough. So there is a salient correlation pattern that WG classifies
as a dependency type and that I classify as a valency pattern. I can't
offhand remember whether I know arguments pro or con the grammar recognizing
valency patterns as categories/concepts in their own right, but probably
there is a need for that.

I am convinced that prototypes play a crucial role in language, e.g. in the
oddball constructions that Joe in particular likes to brandish. But I am
not convinced that prototypes play a crucial role in the general orderly
mechanics of the grammar, if there is a such a thing.

Going back to the original point, though, if the 'Subject' prototype is in
fact a valency pattern then it doesn't count as an argument pro DG.

--And.