Mary Holstege

#balisage DW:

Pros:
Efficiency gains, accelerated development and knowledge transfer, great help system and conversion (but verify), ...

Cons:
Accuracy concerns, IP rights provenance, server energy consumption, ...

#balisage DW:

Example: Using CoPilot to help get familiarity with complex set of DITA customizations

#balisage DW:

The art of the prompt

Need to be succinct, specific, thorough, state the obvious, personae, break complex ideas down

#balisage DW:

How is AI useful to a publisher?

NOT for writing content! (IP, plagiarism, etc.)

But
poofreading
validation of facts and details
summarization
understanding
...

#balisage

How I Stopped Worrying & Learned to Love AI (FP)—Dale Waldt

https://www.balisage.net/Proceedings/vol30/html/Waldt01/BalisageVol30-Waldt01.html

#balisage AF:

Neo4j graph database
BGE-M3 embedding model
FAISS vector indexer (vector search)
Qwen2.7B: the LLM

#balisage EB:

Designing a RAG with TEI

Taking TEI Guidelines p5.xml => digitai-p5.json + query scripts to create knowledge graph (for Neo4j) to instruct local LLM

#balisage AF:

DigitAI a tool to assist scholars and editors to applying TEI guidelines

Small "explainable AI"
Lightweight
Obedient
Multilingual
Interpretable
Modular
Open-Source

Ollama from Meta; missing key languages
Now Qwen2 models Qwen2:7b and Qwen2:7b-instruct

#balisage

Elisa Beshiro-Bondar and Alexander Fisher
Digital AI for Localized TEI/XML Assistance

GitHub - AndrewSales/XQS: XQuery implementation of Schematron XQuery implementation of Schematron. Contribute to AndrewSales/XQS development by creating an account on GitHub.

#balisage AS:

XQS implementation

XQuery for Schematron
Native XQuery implementation of ISO Schematron

XQS version 1.0 released
https://github.com/AndrewSales/XQS

#balisage AS:

Now at Final Draft International Standard

Outline of changes
https://andrewsales.com/schematron4/index.html

#balisage Open Mike

Andrew Sales: Schematron 4 update and implementation

#balisage GB:

Also extension functions to call AI, therefore available in XSLT, XQuery, XProc, etc. etc.

#balisage GB:

DITA for AI prompts

Both XML and Markdown for topic formats; reuse/sharing

#balisage GB:

Oxygen AI Positron: AI plugin for Oxygen tools

#balisage Open Mike

George Bina: Structured Content and XML Technologies in the Context of AI Processing

#balisage DL:

JSON map as argument to call to JS function

Supplying XDM map in ixml:call doesn't work ("reasons")

SaxonJS 3 adds ixsl:json-parse() to construct JSON objects from XSL side

#balisage DL:

Document loading using doc() with request headers

'pool' option for HTTP requests; value
'xml' to add to documentPool so it can be accessed via doc()

#balisage DL:

Combining results of processing multiple source documents: new language features in SaxonJS 3 => cleaner solution

ixsl:promise
@Select using ixsl:all function to parallelize fetches
@on-completion called when done
@on-failure called on failure

#balisage DL:

SaxonJS Tutorial:
https://declarative.amsterdam/program-2021

#balisage

Sponsor Presentation: Saxonica -
SaxonJS 3 coding improvements—Debbie Lockett

#balisage SD:
So... .what about programming languages?

Why? Turn off spell check, don't mess with punctuation, whitespace, syntax highlighting
Where? Same carrier as for NL: should be able to without conflict
How? Standard codes/names would be nice; little regional variation but many versions […]

#balisage SD:
Natural language identification

Why? Get right spelling, punctuation, font...
Where? xml:lang, Markdown ```, file systems extended attributes
How? Standard (ISO) values for language, country, script codes

#balisage

Language Identification for Program Code in Documents (LB)—Steven J. DeRose

https://www.balisage.net/Proceedings/vol30/html/DeRose02/BalisageVol30-DeRose02.html

#balisage HB:
Futures:
More parsers, integration of external tools
Plugin system for user-specific extension functions
Interactive visualization of expression values implemented as HTML pages

#balisage HB:

Definition document includes fall back parsing, e.g. try parse HTML as XML, then fall back to HTML, etc.

So now can navigate seamlessly into various file types

e.g. into CSS and navigate to properties

#balisage HB:

How to get seamlessness?

Need to map file URI => method of parsing

Based on file extension or pattern, folder name pattern (project-specific rules)

=> Infospace definition document

Define resource types
Grammars
Resource type uses (mapping)

#balisage HB:

Addressing limited file formats => Invisible XML

So add functions:
idoc() parse with ixml grammar
iexpand-nodes() expansion of text in tree into element nodes with ixml grammar
ivalid() validate strings against ixml grammar

#balisage HB:
Current limitations

1. Only subset of files can be parsed
2. Explicit call of parse functions distracts from pure navigation

#balisage HB:
Extension functions, e.g.
freq() => frequency distribution with counts
path-content-ec() => all paths in nodes with counts
resolve-link() => from content to file