← News

ResearchSources: arXiv · 16 Nov 2026

Pointer network explains Latin text reuse word by word with 200 annotated pairs

Researchers describe a model that aligns each word of a reusing Latin passage to its source and labels the change, trained with far less annotation than before.

In plain words

When a Latin author borrows from an earlier text, scholars want to know exactly how each borrowed word was changed. The researchers built a model that links every word in the later passage to a word in the source and names the operation, such as a copy or a substitution. It learns from synthetic examples plus a small set of human-corrected pairs.

What was announced

The team says word-level annotation of Latin text reuse is expensive, and only a few corpora annotated this way exist. Their procedure trains a first model on synthetic pairs, created by applying known operations to Latin text, then uses active learning to select pairs for an annotator to correct. A pointer network that links and labels in one step, choosing a source word and naming the operation for each reuse word, performed best among the architectures tested.

According to the paper, 200 pairs chosen by uncertainty through active learning were enough for the model to match one trained on every annotated pair of the 1,490 references in the Loci Similes benchmark. The authors release their code, word-level predictions for all references, and an interactive demo.

What it does not show

  • Evidence is limited to the paper's abstract; the full article was not reviewed, so method details and evaluation caveats cannot be verified.

Sources

Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 17 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.