apostolic-fathers

An effort at a freely-available, openly-licensed edition of the Apostolic Fathers with morphology and lemmas.

Initially focused on Greek, but I eventually want to get the Latin portions annotated as well.

License

This work (data) is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

The code (python scripts) are licensed under the MIT License.

Apostolic Fathers Greek Morphology

Since there is no publicly available and openly licensed morphology of the writings of the Apostolic Fathers (that I know of besides some stuff from OpenText; please correct me if I'm wrong and supply me links), I figured it was time to start on a project.

I'm starting with James Tauber and Seumas MacDonald's Open Apostolic Fathers. This is an edition of Kirsopp Lake's edition of the Apostolic Fathers, and is serviceable for these purposes as it is licensed with a CC-BY-SA 4.0 license.

In typical Rick fashion, I'm starting with the stupidest thing that might work, and that means mooching off of Tauber's other project, MorphGNT, also licensed CC-BY-SA 4.0. Like it is so stupid. I'm just taking forms that happen in the Greek NT (SBLGNT) and migrating the morph data over to the Apostolic Fathers based on (normalized) word matches.

The output currently mimicks the record style of MorphGNT. If I haven't got the value yet, I just insert ?? as appropriate. Note that I've also added a column to the record. The language column, at the end, has valid values of either grc (Greek) or lat (Latin). As discussed below, I hope to run the Latin text through a model to lemmatize and analyze morphology at some point as well.

You can find it here:

• data: https://github.com/RickBrannan/apostolic-fathers/tree/main/data/morph/
• code: https://github.com/RickBrannan/apostolic-fathers/tree/main/code/python/apostolic-fathers/af-morph.py

Current Status

Greek

For the Greek, I'm using this model from HuggingFace, which I've dabbled with before for lemma generation. I'm going to initially use the lemma and morph capabilities, but it has some named entity recognition capabilities as well that I'd like to play around with.

For Greek words unknown by Tauber's MorphGNT, I am now using the grc_proiel_lg form of the model to include lemmas and morph. These are untested and not reviewed. I've done some comparisons of words known by MorphGNT against the morph generated for them and found agreement, but it isn't perfect. Some known areas of trouble include:

• Crasis. The model splits a crasis like καγω into two tokens, whereas MorphGNT treats it as one. These are currently causing some problems.
• Morph mismatches. The model doesn't emit morphology as traditionally thought of by most Koine/Hellenistic Greek Grammars. For example, the model has tense values of past, present, and future along with Aspect values of perfective and imperfective. I'm unsure how well this data can be combined to map over to include as 'tense' stuff like aorist, perfect, imperfect, and pluperfect. But we'll see.

All that said, there is data now available that has some sort of (unproven, untested, and wrong in spots) morphology for almost every available Greek token.

Latin

For the Latin, I'm using Latincy's la_core_web_lg model. The hope is to provide decent lemmas and morphology for the Latin portions found in the writings of the Apostolic Fathers.

I have only initially experimented with this. Some known areas of trouble include:

• Latin POS and Morphology. At present, values returned from the Latin model are mapped using the Greek POS and Greek morphology code. This is subpar and will be addressed at a later date, after the Greek POS and morphology stuff is a bit more stabilized.

I'll document the morphology structure for Latin as it is developed, but I'm guessing it will look much like whatever Latincy spits out, only Tauberized.

Codes and Fields

Columns (Greek and Latin)

• book/chapter/verse
  • unlike MorphGNT, I use three digit fields as the Shepherd has over 100 chapters.
  • unlike MorphGNT, the Shepherd has five values: book/writing/section/chapter/verse. However, I do plan to migrate them to Whittaker's chapter-verse structure.
• part of speech
• parsing code
• text (including punctuation)
• word (with punctuation stripped)
• normalized word (using Tauber's greek_normalisation Python library
• lemma
• language (grc or lat)
• source (MorphGNT or grc_proiel_lg or la_core_web_sm). This is source for both lemma and morphology string.

Part of Speech Code (Greek and Latin)

• A- adjective
• C- conjunction
• D- adverb
• I- interjection
• N- noun
• NP noun, proper (not in Tauber)
• NU number (not in Tauber)
• P- preposition
• RA definite article
• RD demonstrative pronoun
• RI interrogative/indefinite pronoun
• RP personal pronoun
• RR relative pronoun
• TL transliterated (not in Tauber)
• V- verb
• X- particle

Parsing Code (Greek)

• person (1=1st, 2=2nd, 3=3rd)
• tense (P=present, I=imperfect, F=future, A=aorist, X=perfect, Y=pluperfect)
• voice (A=active, M=middle, P=passive)
• mood (I=indicative, D=imperative, S=subjunctive, O=optative, N=infinitive, P=participle)
• case (N=nominative, G=genitive, D=dative, A=accusative, V=vocative)
• number (S=singular, P=plural)
• gender (M=masculine, F=feminine, N=neuter)
• degree (C=comparative, S=superlative)

Disclaimer

Disclaimer: This is totally an in-my-spare-time and as-I-feel-inspired kind of project. And I don't have a lot of spare time. No promises about status and finishing, use at your own risk, etc.