Minutes: MathML semantics meeting Tues, 2 June

Meeting was recorded:
https://benetech.zoom.us/rec/share/9NVEFJzJ2FJIXdLi50XfHbMTOarJX6a8h3UfqaBczEm1bk6lWyYmUtq5nW6970I
      (Access Password: 6z!84$j7)



Attendees:

Neil Soiffer

Moritz Schubotz

David Carlisle

Sam Dooley

David Farmer

Bruce Miller

Patrick Ion

Moritz Schubotz

Big thanks again to David Carlise for taking notes.
BM’s idea of abstracting markup patterns
<https://mathml-refresh.github.io/mathml/docs/layout-semantics>

BM: There has been some rearrangement and expansion of the examples

BM: the document current;y says no patching matching necessary.

BM: To produce such a document you can start from well marked up latex or
content mml etc, but once the keywords were on the nodes you wouldn’t have
to do pattern matching.

SD: But the draft then specifies that this is a pattern that must be matched

BM: added a sentence to clarify this.

BM intention is that the dictionary would compile the #1 #2 etc into xpath
or similar construct to extract the children.

NS: Looking at 2nd to last, stacked fence, what is meaning of <open/>

BM: it is intended to be pseudo-code

BM: in the table I was just trying to list the main composition forms.
Would also have a meaning attribute probably also on the mrow eg
meaning=binomial or meaning=2d-vector etc

NS: you could put a and b for open and close, would you need to validate
the input.

BM: yes when fully fleshed out you could have more validation eg say #1 is
first child of first mfrac

NS: missing 3 most important patterns, infix, prefix and postfix operators.

BM:They could be, but I was thinking that was more or less the default .

SD: but you want to be able to say “factorial of n” (as well as “n
factorial”)

BM: could use a composition like “infix” would work for long mrow with +
and - etc

NS: would be helpful if the document clarified the intended markup for
these, and for function calls

BM: After claiming pattern matching had discussion with Deyan Ginev who
brought up integration. You never know where and how many “d” will be
encountered.

BM: the list of compositions certainly isn’t complete, the question is
whether it’s completable.

SD: a lot of hand waving in the document   eg a+b - c +d could be grouped
in several ways.

NS ({=mrow) can see {a+{b-c}+d} but you can not have {-c} as otherwise
b{-c} would be multiplication

DC: U+2064  is your friend

DC: if the outer mrow was marked as a polynomial the infix - could be
interpreted as prefix - on the term in the outer summation

NS: you should be able to select any subterm the - shouldn’t be bound
closely to the c

SD: you need more analysis than just “infix” or “prefix”

SD: I can’t say whether the attributes are successful as not clear the
intended use.

BM: took the approach that to the extent I understood accessibility was
thinking of whether these attributes were enough. Deferred considerations
of computability to later

NS: I want to separate out the matching part from the template part. In the
accessibility view there is not just one way to say something, depending on
disability and mathematical expertise, plus language differences.  Also CA
systems such as mathematica/maple.

NS: one style is clearspeak could have a template for that and another for
mathspeak based on nemeth braille

NS: There is a pattern, associated with that is a dictionary (tells where
the arguments are), then can be used in the templates for languages.

BM: the table “composition catalogue” each row represents what I am
thinking of as an entry in the dictionary in an “indicative hint” form. The
pattern string wouldn’t be in the dictionary (would use xpath or similar)
the template column is just one indicative example.  You could have
different dictionaries for different languages.

NS: most braille would not use the semantic meanings eg x^T would be marked
as superscript.

SD: there are some cases where the same notation gets different braille
depending on meaning,

MS: interested in Volker’s semantic attribution, hope in wikimedia hope to
expose more accessible mathml

MS: hoping that we could make it swappable to use the speech rule engine
so would like to see a comparison with the speech rule markup

MS: Bruce is there a way to always map back to Content MathML

BM: Sam’s point of view is that there should be a mapping

MS: it would be helpful to include the content mathml representations of
the examples.

DF: I would like to see more examples as needed to disambiguate a calculus
textbook

BM: You have examples added in the issue, I did look at them and check they
could be handled (in theory) but I didn’t add to the document, I should do
that to add more complete examples.

NS: what Bruce is proposing, adding composition and meaning attributes,
then given that you can look up a dictionary which would tell you how to
interpret the children then look in speech based or templates.  Compared to
mathrole, that merges the meaning and composition together.


Meeting next week. I'll send out another poll because this time is not good
for Akashdeep (midnight)