[developers] Dropped arguments in DMRS

[Ann Copestake]

Thanks!

Just picking up on the English case

On 12/01/2016 20:58, Emily M. Bender wrote:
> (ia) I already ate.
> (ib) What? (meaning what did you eat)
>
> (ii) I already dined.
> (iib) #What? (meaning what did you dine on)

but
(iic) What on?
is fine

also:

(iv) I understood.
(ivb) What? (meaning what did you understand) is OK

(v) They blamed me.
(vb) #What?
(vc) What for?

I just refreshed my memory of Fillmore 1986 - I think both blame and 
understand would be cases of DNC (definite null complements)
whereas `read' etc are indefinite null complements (because you can say 
`she's reading without knowing or caring what).

> (iiia) What happened to the cake?
> (iiib) #Kim ate.

Right - but Fillmore attributes that to the fact that eat is an example 
of INC.  A DNC case would be:

What was their reaction to the proposal?
They accepted.

I mean, for interpretation, it would indeed be nice to know that `they 
accepted' => `they accepted something' since that would no doubt help 
with producing the inference:

They accepted the proposal.

but then we'd need to know DNC vs INC.  And I'm not sure it's really 
about dropped arguments:

What was their reaction to the proposal?
They said `not in a billion years'.

one wants to interpret this as a rejection of the proposal.

One other thing - one of Fillmore's main points is that there are some 
differences between closely related word uses as to whether the 
complement may or may not be dropped.  e.g.,

he lost - ok when referring to elections, not to keys
I forgot to fix it / I forgot
I forgot my keys / * I forgot

He argues that this means fine-grained lexical entries are needed. I 
think this might be a nice class of things to experiment with 
distributionally, but, in the mean time, I'm not sure we should be 
trying to handle them in English.

I guess my overall feeling about the nuances of English dropped 
arguments is that this is the sort of thing it would be nice to handle 
well in the ERG, but it looks like the sort of fine-grained lexical 
semantic analysis that doesn't tend to work out when we try and apply it 
beyond a few cute examples.

All best,

Ann