Welcome to ar5iv! Wrestling with the Open Problems of Scholarly Writing

CICM 2022

ar5iv

arXiv

LaTeXML

https://creativecommons.org/licenses/by/4.0/

An invited talk for CICM 2022, announcing the newly created ar5iv platform. ar5iv is an arXiv Labs project, attempting to provide an HTML5 conversion of all arXiv articles with LaTeX sources, via LaTeXML.

2022-09-22

class: center, middle
### Welcome to <a href="https://ar5iv.org" style="margin-left:-6px;padding:0px;"><img width="120px" style="display: inline; padding:0px; margin-bottom:-9px; vertical-align: bottom;" src="https://ar5iv.org/assets/ar5iv.png" title="ar5iv.org"></a>!
<br>
*Wrestling with the Open Problems of Scholarly Writing*
<br>
<br>

.footnote[**Deyan Ginev**]
.remark-venue[CICM 2022<br>
September 22, 2022]

---
class: center
## TLDR: What is ar5iv?

An official preview site for arXiv.org as HTML5 (with MathML)<br>

1.88 million e-print documents<br><br>
97.6% of all LaTeX sources, 76.2% error-free<br><br>
available at [ar5iv.labs.arxiv.org](https://ar5iv.labs.arxiv.org)

---
## Agenda

* <span class="highlight">A 15-year journey</span>

* Some open problems in scholarly writing

* The near future is 80% in sight.

* Public reception / Thank you!

---
## An idea in 2006

Several one-on-one conversations between:
* Michael Kohlhase, KWARC
* Bruce Miller, NIST
* Robert Miner, Design Science

<br>

Common needs:
- "It would be great to modernize arXiv to XHTML."
- "It would be great to have enough MathML for Math Search."
- "It would be great to solve LaTeX conversion to XHTML."

<br>

*Why 2006?* W3C group is re-chartered for work on MathML 3.

---
## LaTeXML

https://dlmf.nist.gov/LaTeXML

* "free, public domain software, which converts LaTeX documents<br> to XML, HTML, EPUB, JATS and TEI" [(wiki)](https://en.wikipedia.org/wiki/LaTeXML)

* Developed for, and maintained by
 - NIST Digital Library of Mathematical Functions
 - [dlmf.nist.gov](https://dlmf.nist.gov/)

* version **0.8.6**, a production-ready Perl application
  * new version coming up shortly after CICM 2022
* implements a variant of the TeX typesetting engine
  * and a small part of the CTAN package ecosystem

---
## A group of enthusiasts in 2007

- arXMLiv project umbrella is created in KWARC
  - I join, age 19, and stick around - age 34 today.
  - together with 15+ Jacobs University undergraduates
  - enable more LaTeX from arXiv by extending LaTeXML
- LaTeXML 0.5.99 released

---
## A first result in 2008
<div class="center">
"Transforming the arXiv to XML"<br><br>
<span style="font-size:80%">Heinrich Stamerjohanns and Michael Kohlhase</span><br><br>
<span class="highlight">CICM 2008</span><br>
</div>
<br>

- 58% success claimed, of ~400,000 documents.

- LaTeXML 0.6.0 released

---
## A second result in 2010
<div class="center">
"Transforming Large Collections of Scientific Publications to XML"<br><br>
<span style="font-size:80%">Heinrich Stamerjohanns, Michael Kohlhase, Deyan Ginev, Catalin David and Bruce Miller</span><br><br>
<span style="font-size:80%">Mathematics in Computer Science (MCS)</span><br>
</div>
<br>

- 61% success claimed, of ~400,000 documents.

- at LaTeXML 0.7.0

---
## Fifteen years in hindsight

---
## Trough of Disillusionment, 2011-2015
 - most arXMLiv members graduated onward
   - I join a startup at 2014
 - available hardware changed every couple of years
 - original build system, now unmaintained
   - a new build system took until 2016 to stabilize
 - incremental progress becomes demotivating
   - each new batch of arXiv sources lowers success rates
 - and yet LaTeXML calmly advances to 0.8.1

---
## Slope of Enlightenment, 2016-2018
 - 2015, latexml introduces OmniBus.cls.ltxml
   - new trend for graceful fallback behaviors
 - The new Rust build system stabilizes
   - leveraging ZeroMQ, PostgreSQL, Redis
   - deployed at [corpora.mathweb.org](https://corpora.mathweb.org)
 - Rebound in conversion success
   - 68.04% success on the original 400,000 docs, up from 61%
   - But! we regress to 59.83% on the full arXiv corpus
       - 1.07 million documents, until 08.2016.
   - return to 64.17% overall success only in 08.2018, latexml 0.8.3

---
## Plateau of productivity, 2018-2022
 - We continue to test LaTeXML with arXiv data every year
   - Bruce tends to add a dozen LaTeX packages per release
 - Michael continues to provide hardware availability at KWARC
 - I start working full-time with Bruce on LaTeXML in 2018.
 - incremental progress becomes motivating.

---
## The vanity slide, 2017-2021
Surprise!
 - arxiv-vanity.org is independently created
   - by Ben Firshman and Andreas Jansson
   - a live preview site for arXiv, initially based on pandoc
   - mobile-friendly responsive design
 - we touch base, vanity switches to LaTeXML
 - arXiv reaches out to all in 2017, interested in having native HTML
   - a number of preliminary steps, arXiv-NG effort
   - ... but internal transition leads to a pause
 - Time for arxiv-vanity maintenace dries up in early 2021

---
## The need for ar5iv, 2021

- new need and opportunity for a (maintained) preview site
   - a lot of community interest was developed by -vanity.
 - In September, I embark on building ar5iv
   - 3 months of heavy CSS work.
   - on good days reviewing 100 articles.
 - we already have the HTML5 data
   - public dataset releases since 2017, part of SIGMathLing
   - https://sigmathling.kwarc.info/resources/arxmliv/
 - sprinkle in some marketing tricks I picked up in NYC
   - take any arXiv URL and replace "X" with "5" to get the HTML
   - ask academic Twitter to spread the word

---
## The ar5iv moment 📢 Jan 31, 2022

<table style="float:left;">
<tr>
<td><img src="/cicm2022/ar5iv_moment_tweet.png" width="350px">
</td><td>People... got excited?</td></tr><tr>
<td>
<img src="/cicm2022/ar5iv_moment_stats.png" width="350px">
</td><td>Half a million recipients in a week.</td></tr></table>

---
## The ar5iv Lab - Feb 21, 2022

Now at [ar5iv.labs.arxiv.org](https://ar5iv.labs.arxiv.org)

---
## Lessons learned
 - only possible with long-term NIST support for LaTeXML
   - and with long-term commitment of core members

- incremental progress is still progress
   - at +2% a year, the 100% in 12 years *(if only...)*

- arXMLiv and ar5iv had **zero** dedicated funding
  - borrowed hardware, borrowed time

- we can not wish away sticky needs
   - "PDF is all you need." - Wrong!
   - arxiv-vanity is proof.

- project burnout is real
    - it takes fresh perspectives to heal

- hype is a force of nature
    - and Twitter is its conduit

---
## Agenda

* A 15-year journey

* <span class="highlight">Some open problems in scholarly writing</span>

* The near future is 80% in sight.

* Public reception / Thank you!

---
## Why isn't LaTeX to HTML solved yet ? (1)

Ten mature conversion tools<br>
 * filter: released pre-2020 with active development post-2020
 * healthy: 10 projects in 8 prog. languages

<table><tbody>
<tr><td>latex2html</td><td>perl</td><td><a href="https://github.com/latex2html/latex2html/">link</a></td></tr>
<tr><td>latexml</td><td>perl</td><td><a href="https://math.nist.gov/~BMiller/LaTeXML/">link</a></td></tr>
<tr><td>lwarp</td><td>TeX, lua</td><td><a href="https://www.ctan.org/pkg/lwarp">link</a></td></tr>
<tr><td>pandoc</td><td>haskell, lua</td><td><a href="https://pandoc.org/">link</a></td></tr>
<tr><td>plastex</td><td>python</td><td><a href="http://plastex.github.io/plastex/">link</a></td></tr>
<tr><td>tex2page</td><td>scheme</td><td><a href="http://ds26gte.github.io/tex2page/index.html">link</a></td></tr>
<tr><td>tex4ht</td><td>TeX, lua</td><td><a href="https://tug.org/tex4ht/">link</a></td></tr>
<tr><td>tralics</td><td>C++</td><td><a href="http://www-sop.inria.fr/marelle/tralics/">link</a></td></tr>
<tr><td>HeVeA</td><td>OCaml</td><td><a href="http://hevea.inria.fr/">link</a></td></tr>
<tr><td>TTH</td><td>C</td><td><a href="http://hutchinson.belmont.ma.us/tth/">link</a></td></tr>
</tbody></table>

---
## Why isn't LaTeX to HTML solved yet ? (2)

<span class="highlight">Answer 1:</span> Of course it is.
 - "Markdown subset" is well-supported by all tools
 - so is basic macro expansion

<span class="highlight">Answer 2:</span> It is impossible.

- a Turing-complete moving target
   - LaTeX 3 is taking over
   - CTAN packages continue to evolve
 - data model mismatch: trees are not boxes

<span class="highlight">Answer 3:</span> Bonsai garden
  - possible for some trees to match with some boxes
  - graceful fallback to layout directives for the rest

---
## Why isn't LaTeX to HTML solved yet ? (3)

The Web is a moving target

*Dark mode papers on smartphones?!*

1. Executable documents
  - reproducible models, interactive visualizations
  - Authorea, Curvenote, JupyterBooks, ...

2. Science movies
  - Arcadia Science is [now using](https://research.arcadiascience.com/pub/resource-microchamber-slide#the-microchamber-apparatus-is-effective-at-confining-cells-for-imaging-over-several-hours) the HTML `<video>` tag

3. Scholarly HTML?
  - schema.org microdata for scientific documents
  - standard markup for frontmatter, figures, tables, listings, calculi ...
      - **interoperable** UI components

4. HTML evolves
  - CSS flexbox, grid, @media
  - MathML 4, ARIA

---
## The LaTeX Social Contract

- arXiv authors write for their human readers
- If the PDF can be read, the job is done
- authors do **not** write to please markup developers
  - we all inevitably take shortcuts
  - the exception proves the rule
- So far, adoption favors messy systems

---
## Markup in a hurry, narrative

1. hand-made abstract (arXiv:2208.XXXXX)
    ```latex
    \maketitle
    \begin{quote}
      \textbf{Abstract}
    \end{quote}
    ```

2. hand-made section (arXiv:hep-ph/95XXXXX)
    ```latex
    \clearpage

{\bf Introduction.}
    ```

3. statements used as list items (arXiv:math/02XXXXX)
    ```latex
    \newtheorem*{intropar}{\addtocounter{thm}{1}\arabic{thm}}
    \begin{intropar} item 1 content \end{intropar}
    \begin{intropar} item 2 content \end{intropar}
    ...
    ```

---
## Markup in a hurry, math mode

1. hand-made footnote marks (arXiv:1012.XXXX)
    ```
      ... Author Name${}^3$ and Other Name${}^4$
    ```

2. hand-made list item decorations (arXiv:hep-ph/95XXXXX)
  ```latex
    $\bullet$ item one content

$\bullet$ item two content
  ```

3. mixed-mode constructions (arXiv:astro-ph/95XXXXX)
  ```
    observed abundances of $^4$He, $^3$He, and D
  ```

---
## Markup of necessity: glyphs

- kerning dots into a diagonal ellipsis (arXiv:1502.XXXXX)
    ```tex
    \def\qdots{\mathinner{
      \mkern1mu\raise1pt\vbox{\kern7pt\hbox{.}}
      \mkern2mu\raise4pt\hbox{.}
      \mkern2mu\raise7pt\hbox{.}\mkern1mu}}
    ```
- old style QED square
    ```latex
    \fbox{\phantom{\rule{.7ex}{.7ex}}}
    ```
    We now have Unicode, U+22F0 (⋰), U+25A1 (□). How can we connect them? <br><br>

- A general problem of connecting absolute positioning to structural counterparts
    - XY/tikz/custom diagrams to SVG
    - hand-crafted title pages to a narrative tree
    - geometric arrangements of custom "mini pages"
    - meaningful borders and lines (QED square, inference rules)

---
## Math expressions are trees
- a layout tree and an operator tree

---
count:false
## Math expressions <s>are</s> have trees
- a layout tree and an operator <s>tree</s> graph.

<span style="float:right;padding-bottom:0.5rem;font-size:50%">(arXiv:hep-th/00XXXXX)</span>
<img src="/cicm2022/iparenj.jpeg" width="680px">
<hr>

<span style="float:right;padding-bottom:0.5rem;font-size:50%">(arXiv:gr-qc/00XXXXX)</span>
<img src="/cicm2022/RC_curvature_tensor.png" width="680px">
<hr>

<span style="float:right;padding-bottom:0.5rem;font-size:50%">(arXiv:1409.XXXX)</span>
<img src="/cicm2022/script_permutations.png" width="680px">

---
## Agenda

* A 15-year journey

* Some open problems in scholarly writing

* <span class="highlight">The near future is 80% in sight.</span>

* Public reception / Thank you!

---
## Track record with conversion (arXMLiv/ar5iv)

* 97.6% of articles can be converted, 76.2% error-free (**+4.1%** since 2021).
<br><br><br>
<div style="margin-left:-55px;"><img src="/sigmathling22/arxmliv_history_2022.svg" width="680px"></div>

---

## LaTeXML over arXiv: what is missing?
**Coverage:** "brutal long tail"
 * 10,100 missing packages and classes
 * 19.1% documents encounter unknown macros
 * 64,500 *distinct* unknown macros, 1.5 million total

**Most needed packages**
 * tikz-cd, epic, biblatex, arydshln, mhchem, tabu, mdframed, eepic
 * about 5% put together

**Math syntax**
 * 1.25 million documents have 1 or more unparsable formulas
 * more than 23 million formulas overall can not yet be parsed.

---
##  ar5iv features

* To be decided: bring your ideas!

* *mini* article viewer
  * inline citations
  * figure zoom
  * margin notes with highlights

* ar5iv footer
   * dark theme
   * "Feeling lucky?", adjacent article navigation
   * conversion log, link to arxiv original

---
class: center,padded-table
## ar5iv examples (1)

---
class: center,padded-table
## ar5iv examples (2)

By language | link
:-- | :--
Bulgarian (section VI) | [1602.06114](https://ar5iv.org/html/1602.06114)
Czech | [1608.03295](https://ar5iv.org/html/1608.03295)
Estonian | [2010.02636](https://ar5iv.org/html/2010.02636)
Finnish | [1702.00277](https://ar5iv.org/html/1702.00277)
French | [1710.09322](https://ar5iv.org/html/1710.09322)
German  | [physics/0512034](https://ar5iv.org/html/physics/0512034)
Hungarian | [hep-ph/0008011](https://ar5iv.org/html/hep-ph/0008011)
Italian | [2105.04227](https://ar5iv.org/html/2105.04227)
Japanese | [2102.10993](https://ar5iv.org/html/2102.10993)
Russian | [1911.02370](https://ar5iv.org/html/1911.02370)
Spanish | [1909.12119](https://ar5iv.org/html/1909.12119)
Swedish | [1303.0939](https://ar5iv.org/html/1303.0939)

---
## Future directions (1)

**ar5iv Goals**
  * "good enough" LaTeXML **coverage**
      * aim: 80% or more (`xy.sty`, `tikz-cd.sty`)
      * we need raw interpretation
  * "good enough" LaTeXML **fidelity**
      * we need a "Scholarly" HTML
      * we need community feedback
  * "**MathML**-native": correct, compact formula trees
      * we need browser support - MathML Core
        * <span class="highlight">Chrome 109, December 2022!</span>
      * we need research breakthroughs in math parsing

---
## Future directions (2)

**More ar5iv Goals**
  * arXiv-native
    * use latest article versions
    * provide preview on submission to authors
  * a modern "Article Viewer"
    * competitive reading experience
    * "get inspired" by eLife, Springer Open, Authorea...
  * strategic auxiliary extensions
    * math search
    * other arXivLabs projects
    * rich NLP applications
      * the world has started making  [ar5iv plugins](https://twitter.com/polymonyrks/status/1489990675903180800)

---
## Future directions (3)

**Ecosystem Goals**

* "Take what we do *well* and do it **great**"
* LaTeXML is "secretely" undergoing a Rust rewrite
  * 40,000 lines of Rust already written, passing **10%** of core tests.
  * Expected between 1-to-2 magnitudes of speedup
  * Ambitious new math grammar embracing ambiguity and underspecification
* We are still not *inviting* enough.
  - More and better documentation
  - More and better error reporting, fallback behavior
* We are still *overwhelmed*.
  - Find simpler data models
  - Do not encourage advanced TeX use
     - Exception: helpful *together with* advanced latexml use

---
## Future (4): Is ar5iv self-obsoleting?

Consider:
- 1990s: LaTeX is best for printed technical documents
- Today: printing has partially faded away
  - LaTeX documents are stuck with PDF (or DVI)
  - because that is the only reliable conversion path
- Soon: reliable LaTeX to HTML (e.g. via latexml)
  - Authors start preferring HTML to PDF
  - PDF reliance fades away
- Once HTML is preferred:
  - CSS for print becomes reliable
  - Modern HTML-authoring languages pick up ([MyST](https://jupyterbook.org/en/stable/reference/cheatsheet.html) is one of many)
  - LaTeX reliance fades away
- 2100?: LaTeX conversion tools are obsolete.

---
## Agenda

* A 15-year journey

* The near future is 80% in sight.

* Some open problems in scholarly writing

* <span class="highlight">Public reception / Thank you!</span>

---
<div style="margin-top:-1rem">Thank you!</div>
<img src="/sigmathling22/ar5iv_joy_1.png" width="780px">

---
<div style="margin-top:-1rem">Thank you!!</div>
<img src="/sigmathling22/ar5iv_joy_2.png" width="780px">