Built a Vector Embedding Database for metadata representations (flattened work and instance) - 20 million vectors can be used to develop a reusable query service for classification and subject headings #bfwe2026

"Pragmatic AI and BIBFRAME" - Matt Miller, Library of Congress.
Recent explorations with goal to develop useful and practical applications for AI/ML that intersect with BIBFRAME such as Vector Embeddings and Entity Reconciliation
#bfwe2026

There's a need to fix subject and value vocabularies in MARC21 prior to conversion and link as many agent authroity records to the bib data in MARC #bfwe2026

MARC21 BFFI conversion via BIBFRAME is being tested at the National Library of Finland - testing the schema conversion pipeline implemented by AI coding agents on a feedback loop. Claude Code succeeded but consumed $1000 of tokens! #bfwe2026

First attempt - clustering and reconciliation via the conversion pipeline to BIBFRAME with mac2bibframe lost precious local context such as classification - LLM could help with agent disambiguation when authority records and bib records had not already been mapped with one another #bfwe2026

"Linked Library Data transformation pipeline using
local LLM to support feature vector methods in work
clustering and authority record reconciliation" - Mikko Vihonen, Nitor Group.
Supporting system vendors to meet Helsinki Metropolitan public libraries requirements for BIBFRAME (BFFI)
#bfwe2026

A MADS/RDF to MARC Authority conversion will be needed to ensure compatibility with libraries still working in a MARC environment and to work with the existing NACO programme #bfwe2026

Long-term there is a need to migrate from MARC Authorities by revising MADS/RDF to have a Native Linked Data ecosystem without reliance on MARC (for Name Authority data). Development is planned in 2027. #bfwe2026

In BIBFRAME this supports display of preferred variants of Names in multiple scripts #bfwe2026

This includes providing variant Name Authority data in vernacular scripts #bfwe2026

In a BIBFRAME environment we want to support multilingualism and internationalisation, without losing any of the data we 'enjoy' in MARC #bfwe2026

English language Cataloguing practices tend to 'transliterate' or 'romanize' non-Latin scripts into MARC fields, then 'transcribe' non-Latin scripts into 880 Alternate Graphic Representation Fields, linked to their corresponding MARC field via the $6 subfield. #bfwe2026

Language encoding is supported in BIBFRAME editor Marva as xml:lang
It can be brought into MARC via subfield $7 #bfwe2026

Library of Congress implemented BCP47 using the shortest language code available and retaining script designations internally, but when the code is published script designation will be inferred. LoC doesn't use regional designations. #bfwe2026

BCP47 generally encodes language as a 2 character code
en = English es = Spanish
But it can be extended to ISO 639 www.iso.org/iso-639-lang...
It can also be used to indicate geographic locale and script #bfwe2026

Kevin Ford, Library of Congress Linked Data Specialist returns to the #bfwe2026 stage with "Language and Script Tags in BIBFRAME and in MARC: a success story"
BCP47 - a standard method to encode language and script information in "language tags" (latest version: 2009) can be used in RDF

As more institutions adopt Linked Data & as AI systems are increasingly used for discovery, could we see the vision of a Universal Library, where a single search provides access to the knowledge of any library in the world, becoming more attainable? #bfwe2026

The plan is to move toward 100% cataloging in
BIBFRAME, connecting UdeC's DSpace-based Thesis Repository with the library's Bibliographic Database. #bfwe2026

MarigoLD has an intuitive interface - it verifies entered information against BIBFRAME requirements ensuring data integrity, and enables cataloguers to see Works and Instances side-by-side #bfwe2026

Making the MARC21 environment more consistent helps with the eventual transition to BIBFRAME.
The FOLIO Linked Data Editor 'MarigoLD' provides a bridge from MARC to Linked Data BIBFRAME #bfwe2026