Preliminary comments on the Draft TC37/SC4 N033

[Tomaz Erjavec]

Dear all,

Lou Burnard writes:
 > Comments very welcome. We have a couple of months to form a TEI
 > consensus about this but I'd like to know as soon as possible if
 > there's anyone out there with an opinion about this stuff!

first, I agree with almost everything Lou has written. Here are two niggles:

>   out that the things (very confusingly) called tags in the matrix
>   representation scheme are equivalent to the ID/IDREF mechanism in
>   XML. It might also explain why the root of an FS in the DAG

Not really equivalent: coreference means structure sharing so there is
no directionality involved like there is with with ID/IDREF; but, yes,
you could model it with symmetric ID/IDREF.

>2. Talking of DAGs, I'm not sure that this mechanism can or should
>   support cyclic graphs. There is a casual reference to these in
>   footnote 3 which I think  needs expansion, or removal.

DAGs are of course by definition not cyclical. And FSs are usually
DAGs; but there have been some suggestions that you could use cyclical
FSs to model some linguistic phenomena; I think Kaspar and Rounds,
Karttunen, and Krieger wrote about cyclic FSs. App. B also has a bit
in cyclic FSs.

As for my comments, the biggest misgiving is what do do with
types/sorts. The TEI has them, but only as attributes; but types also
form hierarchies, and, in some formalisms (like Carpenter) have
associated constraints (introduce attributes). 

As Carpenter etc did not exist when the TEI was written, all this is
right now simply ignored. I think it should be decided whether to 
a) introduce this extra machinery into the DTD or 
b) say in the standard explicitly that type hierarchies and
   constraints on types are not supported by the standard.

Follow comments on the current draft:

Introduction:

ays that the standard is composed of two main parts - the second on 
FSDs is still to come? 

<p>2. Normative references

Lou complains that the TEI is not in the bibliography. But shouldn't
it actually be in this section, as the normative part frequently
refers to it?

Also, why is ISO 3166 (names of countries) needed?

<p>3. Terms and definitions

attribute: I find the discussion here and elsewhere on the status of
"feature" confusing. I think the term feature should be mentioned only
once, and suggested that it be avoided due to its ambiguity.

boxed integer: I think it should say that it marks s.s. in an AVM,
not in a feature-structure.

root: I would delete 'nor any preceding path' as this follows from it
not having any ancestors.

ubsumption: this is the only entry where a math symbol is given - as
they are not used in TEI part I think it would be best just to remove
it (or, otherwise, also introduce it for unification)

type: "some common feature that *classify"

4.4. Shared FSs

"Her *other loves Mary"  "mother" I guess.

"Note that token identity .. does not guarantee the identity of their
values" <- just the opposite, surely!

End of 4:

I would suggest moving Appendix B into end of 4 - it has the same
status and it is strange to have an intro to FSs without mentioning
unification, even though the standard itself does not use it.

5. FS representation (P4)

5.1.Elementary

it seems strange to list all possible values of <f> but not to mention
that a <fs> can be a value as well.

5.4 Symbolic, Numeric, Measurement, Rate and String Values

"rel" attribute is everywhere mistyped as "Rel"

This section is missing a big chunk on <valRange>:

The <sym> element is to be used for the value of a feature when that
feature can have any of a small, finite set of possible values,
representable as character strings. </gap>

<p>5.3 Symbolic...

"This library would have a total of 1620 (3 9 3 5 2 2) entries."
missing &times; !

End: as Lou notes, not only <rate> but <str> has been dropped as well -
I'd suggest putting both in again.

<p>5.5. Singleton,...

Mistakes in P4?

"In a set, items are ordered, and may not be repeated."

urely "not ordered"!?

<p>"The <null> element when used with a feature organised as a singleton
is a semantic error; however, its appearance as a value for such a
feature cannot be flagged by SGML or XML parsers."

I'd imagine it is simple to find such cases using a schema language?

<p>5.7 Boolean, Default and Uncertain Values

"(see definition for a.global)"

this is not provided in the current version - opens the question how
to make hooks into the TEI P4 - as to a (normative!) reference, or
have the standard self-contained (difficult!).

6. Bibliography

I'd suggest adding at least the first HPSG book (Pollard & Sag 98) as
it introduces typed FSs (although quite informally) and set the scene
for the subsequent FS boom.

 Carl J. Pollard and Ivan A. Sag. Information-Based Syntax and
 Semantics. CSLI Lecture Notes: Chicago University Press, Chicago,
 Illinois, 1988.

Best,
Tomaz


-- 
Toma&zcaron; Erjavec         | Dept. of Intelligent Systems E-8
email: tomaz.erjavec at ijs.si  | Jozef Stefan Institute
www:   http://nl.ijs.si/et/  | Jamova 39, SI-1000, Ljubljana
fax:   (+386 1) 4251 038     | Slovenia