[math-on-web] CG meeting minutes, 2018/03/30 and 2018/03/15

Hi everyone,

Today's minutes as well as the previous (which had been very short so I
forgot to post them -- sorry).

The next meeting is April 12, noon (Eastern).

Best,
Peter.


# MathOnWeb CG 2018-03-30

* Present: Moritz, Arno, John
* Regrets: Volker, Dani

* Arno: finding a serializable representation
  * did it because I needed it, get semantics from formula to eval, process
etc
  * looked around for standard representation
  * not so much visual representation (have others)
  * thus the post
* Neil: capturing semantic information in a tree (or else) is a great topic
  * serialization is not important, comes later
  * I'd like to focus on the contents of the tree
* Arno: I agree content is kind but form matters
  * if you need to work on tree that requires a bunch of libs to process,
that does not make it easy
  * something that serializes in a standard way would make it easier for
libraries, to improve exchange between libraries
* Neil: great goal but we haven't really achieved it
  * openmath kind of does that but isn't really usable (outside research
into the format)
  * Arno: agreed
  * parts of openmath community wanted CAS representations that work across
systems
    * didn't work because e.g., mathematica and maple disagree on
implementation
* Arno: today they exchange on the presentation layer
  * e.g., Mathematica and Maple exchange TeX/MathML etc
  * seems like we can do better
  * Neil: true but you can't capture everything
  * Arno: exactly. But if we can exchange something that's better than
nothing
* Neil: they can do something using MathML etc
  * ContentMML and openmath are not rich enough
  * I think you'll have to be careful how you map things out
  * how far do you want to go
  * e.g., once you get to branch cuts, systems do it differently and things
fall apart
  * Arno: falling apart seems extreme point of view
* Neil: some cut off point may be possible (e.g., high school)
  * but people don't seem ready to write
* Arno: in a symbolic system, they want to enter the information
  * but if not, then not.
  * we have a bunch of systems that do things differently, different
heuristics etc
  * those systems are multiplying, they are using their output but not
sharing them at all
* Arno: e.g., my post mentioned MathJS
  * converting to their structure worked ok
  * but most people use a string input, then MathJS does its own parser,
assumptions about invisible times etc, then computation
  * so MathJS has to do that work, every other lib does the same.
  * users can't exchange anything, can't get consistent result
  * but there's
* Peter: I mostly deal with conceptual not procedural but this discussion
reminds me of a related problem in AT
  * heuristics go haywire a lot without a way for authors to fix them
* Neil: heuristics are often 99%
  * semantics would be great but people are not authoring
* Peter: from a conceptual way, you could fix those with interest
  * paving cow paths to improve heuristics would be reasonalbe there
  * not sure about how procedural info could be used that
* Arno: in TeX you have macros etc
  * those are similar and could be leverage
* Peter: I had a discussion about components at a workshop last week
  * remembering many conversations about templating in (MathML) editors,
publishers struggling
  * the web has this of course, not just web components but just regular
components
* Moritz: exposing and transferring extra information is hard
  * example from work with grad students
  * leveragin external information is important
  * this example gets you to wikidata
  * linked data can then take you further
  * however you're using and storing data, use a common dictionary
  * wikidata is a good dictionary, good to use, link and modify
  * http://vmext.wmflabs.org/ASTRenderer
* Moritz: we also analyzed  MathML convertors that can capture semantics
  * few that could (somehow) describe content mathml
  * analyzed them further
  * this project will continue for a few years (since it's Phd projects)
  * accepted paper: https://www.overleaf.com/13063311bcsyrtvqrdry#/50123989/
  * Arno: this looks very interesting
* Peter: linked data should be a major consideration in any of this, also
for subject/content metadata
* Arno: I had only found a few dictionaries
  * open math, proof wiki
    * not sure how related
    * Moritz: not sure; they don't seem connected to any wikibase
* Moritz: the good thing about wikidata is that there are actually users
  * this was rather nice compared to other math projects which usually have
few participants and then the lead left and it died
  * wikidata has thousands of contirubto
  * Peter: background on terms?
  * Moritz: wikidata is displayed via a MediaWiki instance
    * but wikibase is the data repository that drives wikidata (as a data
platform)
* Peter: random question: are SI units represented?
  * Moritz: yes but it's complicated
    * e.g., https://www.wikidata.org/wiki/Q11379 for energy
    * has quality allows you to go past has attribute (which requires
community vote, you'll have to model that on your end)
    * with has quality, you can refer to an item which you can add to
wikidata yourself (anything you can model in a triple store)
* Peter: just a quick word on the workshop since we're out of time
  * https://aimath.org/workshops/upcoming/webmath/
  * many of the CG will be there
  * AT vendors, a11y experts, publishers, actual mathematicians etc
  * should give us a lot of input for future work


# MathOnWeb 2018-03-15

Present: Arno, Peter
Regrets: Neil, Dani, Volker

* Arno: working on abstract syntax tree for mathlive
    * data structure to be parsed and computed on
  * open question as to in how far I could infer semantics from LaTeX
  * putting together a spec, hoping for feeback
  * that might be something interesting for interop
* Peter: mathquill etc were interested early on
  * Arno: Kevin had reached out but was skeptical due to the expressiveness
of LaTeX
    * I'll post a link to the spec after the meeting
* Peter: related: heuristics of AT are a mess
  * I wouldn't even mind if we agreed to them
  * but that would break a lot of existing content
* Arno: different fields have different customs, too
  * Peter: agreed. It seems more likely to handle this via enrichment into
some broader semantic framework
* Peter: I was recently looking at MathJax's PreviewHTML output again
  * it's really quite nice from a pure-CSS perspective
  * and so easy to tweak the output
* Arno: I was surprised how much font-independence I got in mathlive
  * many element styles are constants (e.g., variable margin)
  * http://mathlive.io/?debug=on shows some of the output