Minutes: MathML semantics meeting Wed, 27 May

Meeting was recorded:
https://benetech.zoom.us/rec/share/6OFkcp-z-UVIU4mS9GziY44vDpamaaa8gylKrvcKzEib3hsN2BaG5Hq1BIbdfth0
(Access Password: 3g$NJMp4)

Big thanks to David Carlisle for the minutes!



Attendees:

Neil Soiffer

David Carlisle

Sam Dooley

David Farmer

Bruce Miller

Charles LaPierre

Murray Sargent

Steve Noble

Akashdeep Bansal

Patrick Ion

BM’s idea of abstracting markup patterns
<https://github.com/mathml-refresh/mathml/issues/217>

NS: Bruce to introduce alternative markup to mathrole.

BM: I don’t know too much about accessibility but started with simple idea
of putting meaning on tokens doesn’t cover all cases eg powers v binomials
v …

Where there isn’t an obvious token where you would put the meaning?

Sam’s idea of putting the meaning on the structural element makes sense in
these cases

But need lots of special cases eg different notations for binomial coef

This seemed to be mixing up semantic markup with the markup structure.

Teasing these apart…


Superscript used in different ways. Powers, operators (eg transpose) and
indices (tensors,...)

And sometimes just used as a unique variable.

Used “composition” as the name for this concept in the notes but may not be
the best name

For transpose: superscript is an operator but put the meaning “transpose”
on the T, which can be read as “transpose of …” the reading pattern (by
default) would work for a large range of cases.

This points to a benefit that we have a small set of composition patterns
which indicates which of the children are “arguments” to the operation,
which works for a large set of meanings on the tokens.

For more complicated markup like binomial, has fences a frac (with zero
line)

Again place the composition “stacked fence” on the outer mrow. Will work
for other operators using the same basic layout.

SD: “basically work” you mean you can produce a vocalization of the math
but would not work for Braille

BM: hand waving: it may fail in details but should work, a dictionary of
composition would have a specification of where the arguments are and a
vocalization paren eg “#1 of #2 and #3”  but that is English vocalization
could have different language and braille etc

MS: Braille has fewer problems as it very closely follows the syntax

SD: still need to know which symbols go in what pattern

MS: Nemeth braille superscripts just follows superscript levels, do not
follow meaning

Nemeth himself preferred it read as sup, supsup etc not “power”

SD: happy to see two attributes one for meaning and one for style
(composition)

BM: Main idea is to disentangle the “meaning” and the “arrangement of the
arguments”

NS: example of exponential exp(...) v e^{...}

SD: previously I was assuming most information would be covered by mathrole
with class for special cases, but this has more analysis of the
presentation constructs

SD: This could be combined with David Farmer’s classification of textbook
notations

SD: then we don’t need to be so strict on which system is used

DF: but we don’t want multiple ways to mark up same thing.

NS: most editors don’t group mrows so in mrows, the * is probably in same
mrow as the + so need to parse the mrow first.

DC: requiring good mrow structure for well marked up mathml doesn’t seem do
bad (most mathml won’t have the attributes at all)

PI: we are proposing a new intermediate markup language  (via attributes on
mrows)

SD: the structural patterns closely related to xslt mapping from content to
presentation

BM: idea is if the mathml you receive is marked in this way, you don’t have
to do pattern matching to infer meaning. If it hasn’t been marked up this
way then adding it does need the parsing and heuristics which should then
generate the annotated mathml.

BM: current proposal isn’t addressing how to add these attributes.

NS: two things to think about

1. in near term likely to hand authored,

2.  in longer term hopeful machine generated, how much work is it to
machine generate.

BM: sam has a tool that works with Cmml internally and should have data to
generate this, in the cases where latexml can figure out the meaning it
could generate this.

DC: any system converting to this from cmml or openmath etc should be able
to generate any markup of this nature, hard case is hand authoring (which
is why not much cmml or openmath ever produced)

NS: no hope that the majority of pmml will have any of these attributes. So
have to assume some defaults. Eg default msup to be composition=power if it
doesn’t have one.  May need some pattern matching for setting defaults --
a+b and f(x) are different patterns but should default..

BM: didn’t think too much about defaults, initially thinking about minimal
markup that would allow full specification, need to think more about
defaults. But would get to the point where if you haven’t been given a
sufficiently annotated markup you have to fall back on parsing and guesswork

NS: have to have guesses there but gives a way to allow AT to avoid
guesswork.

DC: gives a way to serialize the guessed semantics could have a tool that
produces fully annotated mathml (using some implemented heuristics) rather
than it being hidden  as mathml is included into an application.

DF: More optimistic that this could be used fairly quickly, looking at the
two books I surveyed

DF: two ways you can markup for meaning, outermost or innermost.

SD: Outermost is likely to be easier.

PI: pretext or similar system would provide this markup.

NS: Bruce and David F had some experiments with generating annotated mathml
from latex using a customization of latexml

BM: yes long term research with how much you need to make the author do and
how much you can infer from generic latex markup

NS: do people think that this is a good avenue to pursue?

DC: I’m initially in favour

NS: I have a list of special cases from mathplayer. I’d like to create a
bigger document with more examples. Could use google doc or the markdown
(which allows texify math)

Bruce could you start using the same setup as the document Sam and Neil
started (doc folder of mathml-refresh/docs

DC: takes action to set up mathjax on the gh pages documents.

BM: I’ll get something started.

DC: I can smell dinner

MS: want user-settable defaults.

NS: I have joined a different W3C group which meets at this time so will
make a doodle poll for a new time


<http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
Virus-free.
www.avg.com
<http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
<#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>