[Prev][Next][Index][Thread]
Re: the RM saga
> > > Well, in correct DIPETT trees, PPs don't modify each others' objects.
A correct DIPETT tree *could* have PPs modifying each others' objects.
Since the object of a PP is (usually) an entity() structure, any valid
entity() arguments are allowed, including np_postmodifiers() with PPs.
The fact that we don't ever see them is purely a function of DIPETT's
current biases (in this case, the conscious decision to prefer flatter
lists of PPs).
> > I thought this was the whole point of reattachment: to be able to
> > move PPs from modifying verbs to modifying nouns (wherever the noun
> > might be). I can't imagine what use reattachment would be without
> > this.
>
> That obfuscates a bit. I suggest that flat and hierarchical NP post-
> modification are both legitimate treatments of 'the printer on the desk
> in the office at the end of the hall', and that DIPETT at the moment
> with its conj_pps(implicit_and) structure is predisposed to the flat rep.
There's at least one PP in there that can't be attached at the top
(or even second) level: 'of the hall'. My point was that I don't
see the use of changing:
I kicked
the printer /* (I kicked) the printer */
on the desk /* (I kicked) on the desk */
in the office /* (I kicked) in the office */
at the end /* ??(I kicked) at the end */
of the hall /* ??(I kicked) of the hall */
to:
I kicked
the printer /* (I kicked) the printer */
on the desk /* (the printer) on the desk */
in the office /* (the printer) in the office */
at the end /* ??(the printer) at the end */
of the hall /* ??(the printer) of the hall */
> This might be a good point to introduce a theme that was brought up a
> year ago and which Sylvain touched on earlier today--correctness. As I
> recall, people last year thought that the reattacher should not be able
> to produce wrong trees, ie the functionality to do so not be implemented.
> I think we need to decide whether we will allow PPs to modify each other
> to indefinite depth or not.
But PPs modifying each other to indefinite depth is allowed by DIPETT.
> Tokens were put in only to facilitate user interaction. Will you be
> interacting with the user to the point where you'll need to show phrases?
Well, I already use clause tokens in CLRA (if they're available). And
I can definitely envision showing the user PP tokens. Tokens within
the NP proper (without np_postmodifiers) are redundant to me, since
I'm already stripping out each individual word (typically:
intensifiers, determinatives, attributes, noun_modifiers and head).
Ken
Follow-Ups: