[matrix] Message-free Matrix

[Dan Flickinger]

> I have a question on the iforce types:
> ...
> I see the ERG uses ne_x_rel for tag questions, as described in ...

Just a quick note of explanation about that ne_x_rel in the ERG.  Since
tag questions in English employ auxiliary verbs and pronouns, and since
it seems best to use the ordinary auxiliary and pronoun lexical entries
for this construction, we end up with a semantics for 
"Dogs can bark, can't they?"
where the MRS contains an EP for each of the occurrences of the
auxiliary verb (and also for the pronoun).  (The presence of these EPs
is due to our basic assumption about monotonicity in semantic composition
in the Matrix - once an EP is introduced by some lexical entry or by a
rule, it cannot disappear in later composition.)  In order to make this
semantics well-formed, there needs to be a two-place relation much
like that for conjunction, which takes as arguments the top handles of
the main clause and the tag question "clause" (modulo qeqs as usual).
The ne_x_rel is the predicate introduced for that two-place top-level
relation.  

I doubt that there is motivation for such complex semantics in other
languages, which much more typically use a single particle like the "ne"
in Japanese, in place of the English tag question.  A closer analogue
in (American) English would be the "right?" "Dogs can bark, right?"
Here I think it would be nice to agree on a harmonized representation,
maybe along the lines Francis proposes, with some further specialization
of the iforce subtype.  It is hard to decide, at least for English,
what single value to assign to SF, very much like the difficulty in
deciding for a coordinated sentence like the following:
  "Chiang arrived, but will she stay?"
It's certainly not simply a statement, but it's also more complicated
than just being a question.  Maybe the value of the SF attribute for
such examples, and also for the outermost SF value for sentences with
"right?" or "ne" tag questions, should be something like [SF conj-sf],
an explicit indication that there are distinct component SF values
contained in the full sentence.  Then the SF value for "ne" or "right?"
could well be 'ne-ques'.  But in the interest of cross-grammar harmony,
I think it would be better to reserve the type 'ques' for the real
questions, since this type will show up in actual MRSs as the value of
SF.  Then we would just need a good name for the new immediate supertype
of 'ques' and 'ne-ques' - maybe 'abstr-ques'?

  Dan