Minutes: MathML intent meeting, 12, Nov 2020

The meeting was recorded:

https://us02web.zoom.us/rec/share/HOKKm0COiXrbccpMuB16XuAxN7b4QWuLOO4emXF8WyJX_JLE3OdyWUfYC68k23ZN.oUiK4GmXjjTVKBs6
<https://us02web.zoom.us/rec/share/HOKKm0COiXrbccpMuB16XuAxN7b4QWuLOO4emXF8WyJX_JLE3OdyWUfYC68k23ZN.oUiK4GmXjjTVKBs6>

 Passcode: 5*tq%Pvx

Attendees:

Neil Soiffer

Louis Maher

Moritz Schubotz

Bruce Miller

David Carlisle

Deyan Ginev

Steve Noble

Murray Sargent

David Farmer

Sam Dooley

David Farmer

Patrick Ion

Matthew Scroggins

Thanks to Louis Maher for taking the minutes.


1. Moritz to demo his system that uses Content MathML

[sharing his screen vmext.wmflabs.org/ast-render.thml -- presentation
MathML, content MathML, and a tree showing E=mc^2]

Screen shot:


If you hover over a node in the tree, a tooltip show more info. It grabs
info from Wikidata because the content MathML the content is the Wikidata
id. The presentation MathML is the ‘E’, etc. Using ‘E’ in the content
MathML gives no new info, so the Wikidata info is more useful and ‘E’ is
not important to semantics. The content and presentation are crossed linked
via id and xref.

The tree is editable and it reflects in the content and presentation on the
left of the window.

DC: normally the presentation and leaf content is the same. What you do
works for your app, but I’m worried about other scenarios because if you
generate presentation from content, you’d normally use the leaf data.

SD: Suggested that if the intent attributes were there, you would have a
simpler issue.

Mor: He fetched information from the content dictionary in Wiki data.

LM: Seems susceptible to web problems

SD: you could fetch and precompute things for your document so it would not
be so dependent on fast connectivity.

NS: What is the advantage of going to two trees vs one tree with extra
attrs?

Mor: The content mathml is a tree on its own and it is easy to visualize.
With two trees, it seems easier to see. Theoretically, they are the same,
but with attrs, you need to parse the attr string to get the content tree.
So more effort.

DC: The content MathML was generated at same time as presentation. But just
as the content tree was generated from TeX, it could be generated from
MathML.

MURR:

 For speech, speech is closer to presentation MathML than to the Content
MathML.

NS: Symbols can change their meaning dependent on context.

MOR: Wanted to show gold standard content MathML. Wanted that to train
machine learning algorithms. Content MathML difficult to be done by hand.
He developed the tool to edit problems in the content tree. If there is an
API to transform augmented presentation to a content tree, that is
something could use instead.

DC: The syntax can be too complicated to be read by people easily.

NS: We get pushback on embedding a syntax into intent.

DG: SVG has a complicated “transform” attr, so there is precedent.

NS: It’s there in CSS also. Not sure if people think of it as a good idea
or an aberration with historical roots that should be avoided. We will have
to reach out and ask.

DG: I like the concept of a Wikidata CD. But everything in the world is in
one CD, which is not good. That’s why I’ve been working on level 3, which
can map to wikidata but has a readable name. A toolchain question of
whether you prefer XML or annotations? Both should be a simple as possible.

2. Sam to present his ideas on 'intent' (especially implicit arguments).
See https://mathml-refresh.github.io/mathml/docs/intent.html
<https://mathml-refresh.github.io/mathml/docs/intent.html>

Potentially we will examine some of the examples with Deyan's live demo
site:

https://dginev.github.io/tiny-mathml-a11y-demo/


SD: Intent creates micro syntax. Maybe this will result in pushback from
W3C members.

My idea is to come up with defaults so intent doesn’t need to be used much.

Sort of like aria -- it is much better to use the standard rules than to
include ARIA.

NS: I’ve been told many times that one should avoid aria when there is an
HTML way of doing something. Eg., use <h1> rather than <div role="heading"
aria-level="1">

SD: The goal is to specify defaults that work most of the time. These are
hints for how to interpret something.

Intent can take a number of different forms.

I’l use ‘@’ to mean ‘apply’ as in intent=”@plus” means ‘apply plus to its
arguments’ which are implicit in this case.

I’ll use ‘$’ to refer to numbered or named arguments (arg=’...’).

Syntax of ‘intent’ can be a numeral, a name, a variable or function
symbol.  Intent can be a name followed by a list of comma separated
arguments. Typically you would only need to specify the arguments in
unusual situations

<mi> and <mn> have implicit intent of their own contents

<mo> is more complicated as speech probably needs to translate the various
symbols to names and so there probably needs to be a dictionary somewhere
that supplies that translation.

SD: wants to talk about how to get the value out of the Intent attribute.

<msup>is usually exponentiation; need to use ‘intent’ if it isn’t
exponentiation

Wants default rules to make things simpler.

NS: You didn’t mention mrow. What the defaulting rules for that? Are they
an implicit argument like an <mi>.

SD: I haven’t worked that out yet.

NS: Same for msqrt, mfrac, etc.

NS: For next week discuss when the defaults are valid.

SD: will try to match MOR:’s example.

People should send SD: cases of what they want to see.

DG: His system will grow with examples. FYI, I’m up to ‘n’ on my level 3
classification.