Internal deadline:

October 1 -- create document that we are happy to distribute to
Petr's class, and folks like Lars, JP...  Document should be consistent
with Matrix v0.5 (0.5.1?) which should be consistent with new
insights gained from ERG.  Flag open research issues, although
we may not get to all of the prose outlining (in footnotes) the
different possible angles to take on those issues.

Some questions:

[#1] Why exactly do we expect not to need features on semsorts?
(Or ]are we just taking that position to start with because it
is more constrained?)

This is a hypothesis that the system is simple in certain respects,
so we are going to simpler structures until the need for more
complexity is shown.  We don't expect to need reentrancy on some
subpart of sortal information, and so features aren't motivated theoretically.
Likewise, the hierarchy is relatively simple for now, so there
appears to be no engineering reasons to add features.

[#2] Are there potentially cases where a verb might constrain
both the CONT.HOOK.INDEX and the CAT.KEYS.KEY value?  If so,
they would be interesting to discuss.  

None so far.  Maybe discuss cases where each is used separately,
and raise issue of finding a case that uses the distinction.
(Discuss why the syn/sem head path difference isn't really
exploited in the ERG, quantifiers passing up noun's index and
all that.)

[#3] I adapted the example of abstain_v1 to be consistent with 
the placement of --COMPKEY in LKEYS (not KEYS) and with 
--COMPKEY being sort valued.... or wanted to, but didn't know how
that sort was going to be spelled.  Dan, can you update this?

Dan will update once the question of two sortal hiearchies (predsort
and semsort) or one is settled in the ERG.  

[#4] In 4.1 we say that nouns and verbs identify their LKEYS.--KEYREL.ARG0
with their CONT.HOOK.INDEX, but we don't say what happens for prepositions,
adjectives, and adverbs.  What should happen?

adjectives and prepositions as modifiers don't have ARG0.  Construction
to make predicatives out of them creates new relation with reentrant
PRED value and [ARG0 event].  A second construction (reduced relatives,
generalized) makes postnominal PP modifiers ["The grammar currently
in CVS"] out of the predicative ones.  Same for postnominal AP modifiers?

[This is clearly a case of open research questions, to be flagged.
Invite input from other language grammars.]

Illustrate semantics of modifier PPs and predicative ones, so people
know what their grammars should look like.

We're worried about cases like "the currently unavailable solution",
but might be able to get around this by making all such adverbs
scopal.  Expain this solution and flag as working hypothesis.

If possible, it would be nice to dig up again the PP modifying
adverbs that led us to abandon the scopal modifier solution in
that case, too.

Adverbs never have ARG0 or anything but ARG1.

[#5] Are we going to specify PRED values for things like named_rel,
in addition to the distinguished relation types?  If so, we should
be consistent and do this every place we don't expect there to be
variation in the PRED value of a type (cf. message types).

Always have PRED values, explain MRS machinery reason for this.
Also explain new naming convention: _relation for types,
_rel for sorts/strings.

[#6] v0.5 doesn't have any types like noun_arg1_rel, although these
are mentioned in the paper.  Fix v0.5 or put a fn in the paper? 

Put in matrix.

[#7] Also, in the quick mention of deverbal nouns, should we say something
about the analysis of nominalizations (i.e., a nominalization_rel
wrapped around the event_rel) in the brief mention of deverbal
nouns?

Yes.  Especially because the anaylsis illustrates one (some?) of
the special relation types.  The mention of deverbal nouns might
not be the right place for it.  Maybe under special relation types?
There can just be a pointer from the mention of deverbal nouns
to wherever it ends up...

[#8] I'm not sure I'm convinced about temporally constrainted NPs
being interpreted outside the grammar...  Maybe we can just be
a bit more vague and relegate this to future work?

[#9] The paragraph describing the HOOK.INDEX values of adjectives
and related issues was pretty dense.  I'm not up to fixing it right
now though.

[#10] Should we mention the whole reduced relatives issue for postmodifiers
of Ns in English?  (I.e., the idea that we came to this summer about
having modifier PPs only modify verbal projections, and having 
nominal PP modifiers actually be reduced relatives, and include
a proposition...)

[#11] Would adverbs ever have an ARG1 value?  Also, I'm a little unclear
as to why "with care" gets its own ARG0 event variable while "carefully"
doesn't...

Resolved.  See [#4] above.

[#12] The names of the message types (at least the leaf types) need
to be fixed in the paper and in matrix.tdl in order to be made 
consistent with the naming conventions.

[#13] Give specific page ref in G&S 2000 for questions embedding propositions?

[#14] This, from 4.2.3, is inconsistent with what we say in 3.5:

[DAN: Why do we need a separate type here?  Isn't being the value of a
{\sc c-arg} distinctive enough? [DPF] No, it turns out that the
grammar needs to be able to exclude conjoined phrases in certain
contexts, and the only obvious distinction available is this conj-ind
type, since the syntax of a conjoined phrase is by hypothesis not
distinguishable from ordinary phrases.  We could maybe do it with the
(new) SORT attribute instead, but will have to wait and see.]

One or the other needs to be fixed...

[#15] The last paragraph of 4.2.3 re L-HNDL and R-HNDL in coordinated
NPs is opaque to me.  Wouldn't the value of these features at least
qeq the LTOP of the two noun phrases?  If not, why not?

Add a couple of sentences reminding readers that the approach to
underspecification means that quantifiers' handles are never subject
to constraints (although there are some constraints from the 
constraints on bound variables).  

[#16] I fixed the mrs for the office char arrived to match the avm
macros we've otherwise been using, and also put in the standardized
PRED names. But I wasn't sure: should _def_q_rel have a leading
underscore?  It doesn't in the ERG at present.  

[#17] What's the regularized form of n-n-cmpnd?  (I've put in 
n-n-cmpd\_rel in a few places.)  Is it going to be a special relation,
or just a PRED value on a prep_mod_rel?

[#18] Also, should we footnote the display of PRED names as types
as a display thing, regularize to PRED values (and show the
corresponding relation types) or leave as is?

We should be explicit about when we're displaying an avm and
when we're displaying an mrs.  Describe in prose what the conceptual
and typographical differences are, and refer to what people will
see in the LKB.  oe sounded inspired to write this bit...

[#19] In the discussion of linking types in the ERG, I replaced the fake
type names with real ones, but then changed the constraints around a
bit to anticipate --KEYREL instead of relation-valued KEYS.KEY.
Should we add anything here about qeqs in the two scopal types (insist
and probably)? [Didn't quite get to finding the right type to replace
subj-rais-verb, obj-rais-verb, or expl-subj-verb with, also, my choice
of types and placement of footnotes may be somewhat inconsistent.]


[#20] 6.8.1 insert ref to Levine (? -- whoever Ivan keeps refering
to regarding that other theory of XARG) or possibly Ivan's French
rel cl paper (need permission first).

Also describe how our approach differs.

[#21] I didn't see the point of the example "The dog believes there
to be a cat" -- we should either spell this out more, or leave it
out, I think.

[#22] Sem-I:  I'm still not convinced that matrix v. language specific
and automatically generated v. hand-built are really going to line
up.  I think we should probably say less rather than more here...

Something on the order of what we put in the slides for the ESSLLI
talk, just enough to keep people from elaborating different role
names, and to point to future work.

[#23] Sections 7 and 8, see notes in .tex source.  Give pointers to
specific sections of Copestake 2002 where possible & relevant.  Flag
likelihood of changes in the technology?

