@emchateau thanks for the hint to David J. Birnbaum's collection!

Funny coincidence, I am just on my way to what will surely be a wonderful and productive workshop on #Collation and #Visualisation (https://www.lorentzcenter.nl/seeing-the-difference-visualizing-textual-variation.html), and am reading (and can recommend) a #Balisage paper by said D. Birnbaum & Ronald Dekker in preparation: https://doi.org/10.4242/BalisageVol29.Birnbaum01

I will check out the #XSLT learning resources you mentioned and the ones other might still mention during the week, thanks again!

CC @petrichor

#balisage SB: XML corpus analysis, e.g. looking for inconsistencies, looking for choices made when there are multiple ways

Poking around in XML corpous:
Regular expressions: ubiquitous, east to learn, widely useful, powerful
XPath: extraordinarily powerful, only use case is XML (mostly), finding info is not hard, but harder than for regexes

#balisage SB: "If you only knew the power of XPath..."

Taking about the part of XPath that is for tree navigation
XPath has developed and gained a lot of power, but basic tree navigation is the same

#balisage SD: Brief overview of natural language history of LLMs

probability and language => hidden Markov models => word embeddings => attentions => transformers

#balisage MSM: Get there one step at a time
1. Get Word to rudimentary XML
2. Recognize basic record/field structure
3. Recognize 'macro' structure (groupings of fields)
4. Recognize fine-grained structures (e.g. bibliographic sources)

#balisage MSM: Context is information about Roman trials, but it comes from other sources e.g. allusions from prose; create record of the trial with as much who/when/what as can be gleaned. Lots of missing and uncertain information (which needs annotation). Variable structure.

Problem: how to get from Word => 'natural' XML representation focused on meaning not layout

Constraints: simple enough for one man band, resist second-system syndrome