Re: Active again.

Hi!

Daniel Andreasson Vpc-Work <daniel.andreasson@...> writes:
...
> She:AGT hit him:PAT and fell down:DIFF.
Ah!  Is that not a change of voice?

> This means that even though the left out argument of "fell down"
> SHOULD be a PAT and thus "him", it's marked by "DIFFERENT"
> and then you know that it is in fact "she" who falls down.
Hmm.  Or is that only allowed when referring to another verb in a
previous coordinated clause?

> So, leading to your question with two one-place predicates,
> where one is AGT and the other one is PAT. This would be:
>
> She:AGT ran and fell down:DIFF
Hmhmm.  That's currently impossible in Tyl Sjok, too, since the
agent->patient switch cannot be marked there.

> Some sort of voice, like Henrik suggested, would perhaps be
> possible. I'm eager to see what you, Henrik, can come up with!
Actually, by solution I was thinking about was essentially the same, I
just regarded it as a voice.  There could be a particle, say VP (for
'patient voice') that switches a verbs valence from taking an agent to
taking a patient:

    (fall man.PAT)  ==  (man.AGT VP.fall)

(AGT and PAT are not markers, but indicated by word order only)

Then you could compose:

    she.AGT (ran AND VP.fall)

as a short-cut for:

    (she.AGT ran) AND ((she.AGT) VP.fall)

Note that in that construction, the 'AND' links two heads that both
get 'she' as their agent supplement.

The problem I'm having in Tyl Sjok is that I could not find a nice way
to implement this without violating the principle that 'full words',
the only open lexical class I have, do not have a valence in that
sense.  When you compose A and B, it is only clear that A controls B.
It is not clear, however, what is the supplement to what.  So which
one is the head is, well, a change in focus.  Strange...  Therefore,
the compositions agent+verb and verb+patient are equivalent, namely
both are controller+controlled.  And therefore, it's hard to define
which item a voice particle should modify.  The only way it could work
would be by modifying both.

Thus, I though about a different composition method that switches
controller/controlled, say, SWAP.  Then:

   a b == b SWAP a

But, is that a voice?  Maybe 'swapped voice'. :-)

Usage:

   (fall man)  == (man SWAP fall)

and further:

       (she ran) AND (fall she)
   ==  (she ran) AND (she SWAP fall)
   ==? (she ran) AND (_ SWAP fall)

Here, _ is an empty node that is coindexed with 'she'.

The problem is, that 'AND' still connects phrases, so I'm not sure
whether that helps much and whether that really means what it should
to a person speaking Tyl Sjok.  (If I introduce empty nodes anyway, I
could simply define that they may be coindexed even if one is AGT, the
other PAT.  Well.)

Or should I define that it could be interpreted with the following
structure:

   ==? she (ran AND (SWAP fall))

Hmm...

Oh, and although this looks the same as above, it's different:

   'She falls and calls for help.'

   (fall she.PAT) and (she.AGT (call help.PAT))

Short:
       (fall she) and (she (call help))
   ==  (fall she) and ((call help) SWAP she)
   ==? (fall she) and ((call help) SWAP _)

(Or:
   ==? ((fall and ((call help) SWAP)) she)
)

But I was not satisfied with that very much.  If it ever gets
implemented, then probably similar to that.  (It would allow for heavy
NP shifts, too, which are sometimes *very* useful in an embedding
language with deeply nested clauses...).

**Henrik