Citing Computational Assistance

Citing computational assistance

If pyaegean helped produce a result in your work, cite what it used, so a reader can trace the claim: the edition or dataset, the tool and its version, the exact number, and any place a human corrected the automated output. This page shows how. It is a how-to for citing your own work, not a methods paper about pyaegean.

The guiding idea matches the rest of the toolkit: name the register. An established fact, a measured number, and an exploratory reading are three different kinds of claim, and a citation should make clear which one you are leaning on.

Cite the corpus (or the exact subset)

Every corpus carries its provenance (source edition, license, citation), and any filtered subset carries a note recording the filter, so you cite precisely what you used:

from aegean import load
c = load("isicily")
print(c.cite())                       # a ready citation line for the whole corpus
print(c.provenance.bibtex())          # BibTeX
print(c.provenance.apa())             # APA

sub = c.filter(site="Catina")
print(sub.cite())                     # cites the corpus AND records "subset: filter(site='Catina')"

A query result cites exactly the query behind it (QueryResults.cite()), so a table in a paper can point at the precise slice of the corpus that produced it. Underlying editions keep their own rights: cite the original edition too (each document's provenance names it).

Cite the tool and its version

Reproducibility needs the version. aegean.__version__ and the repository's CITATION.cff (GitHub renders it as a "Cite this repository" button giving APA and BibTeX) identify the release. Pin the exact version in a paper so the numbers are reproducible:

import aegean
print(aegean.__version__)

Cite a measured number reproducibly

Measured claims come from a recorded protocol (Methodology, Benchmarks). For a number you report, an evaluation receipt (aegean.greek.eval_receipt) is a content-addressed record of the settings that produced it, so the figure in your paper can be checked. Cite the number with its protocol, not on its own.

Cite reviewed (human-corrected) output

If you exported machine annotations, corrected them, and used the corrected corpus (see When the Tool Is Wrong), that corpus records the correction: each corrected token keeps the machine value under a <field>__pred key, and the provenance gains a review: note. Say so in your methods: which fields were machine-produced, that a human reviewed them, and who. That is more honest than presenting corrected output as if it were either fully automated or fully hand-done.

Cite exploratory output as exploratory

Exploratory results (Linear A / Cypro-Minoan structure, AI readings, generative translation) are labeled unverified at the point of use, and a citation should carry that label: present them as hypotheses generated with computational assistance, not as readings. Do not let a formatted table launder an exploratory result into an established one.

A worked phrasing

Inscriptions were read from I.Sicily (Prag et al., CC BY 4.0) as distributed in pyaegean vX.Y.Z; lemma and part-of-speech annotations were produced by pyaegean's neural joint pipeline and reviewed by the author, with corrections recorded in the corpus provenance; the reported lemma accuracy is the UD-Perseus test figure from the project's benchmark protocol.

See also

• Data & Provenance: what provenance records and how to pin a data snapshot.
• For Specialists: the register model in full.
• Benchmarks and Methodology: the measured numbers and the protocol.