@diyclassics
Associate Research Scholar, Digital Projects @ ISAW Library | prev. Quantitative Criticism Lab (UT-Austin) & Culture, Cognition, Coevolution Lab (Harvard) | Fordham PhD, #Classics | #DigiClass, #Latin, #Greek | LatinCy dev, CLTK contrib | #Python
Side-by-side conllu file in plaintext vs. formatted preview on an example sentence
Wanted an easier way to preview CONLLU files in vscode, couldn't find one, worked up my own...
Building out this kind of ad-hoc tooling—relatively quickly!—is a clear codemodel pro. And this isn't even ad-hoc anymore, already part of my daily workflow
Would love to get some feedback on this—still, not on the marketplace yet, definitely beta, build-from-source, use-at-your-own-risk, etc. etc. github.com/diyclassics/...
Side-by-side conllu file in plaintext vs. formatted preview on an example sentence
Wanted an easier way to preview CONLLU files in vscode, couldn't find one, worked up my own...
Coming up tomorrow! Really looking forward to this...
🎂 A successful first dev meeting in the books—discussion about doing it again for September 2026, details soon. 🎂
latin reader ui with an excerpt from boethius open showing highlighted word pendentis (supra verticem gladii) glossed and parsed in a side bar
looking forward to checking this out! i have been using LatinCy tagging for a latin reader UI i have been working on and performance has been impressive
Maybe we can figure out a way to align such an AG grammar more closely with model dev—happy to talk, run experiments, etc.
Here's another model for today's meeting...
Ancient Greek LatinCy lg model, same workflow/infrastructure at the Latin models, obv. different texts/data!
Experimental, beta—but on HF now: huggingface.co/latincy/grc_...
#digiclass #nlproc
All day today—starting in 10 minutes with the State of the Project report...
Comparative metrics for different LatinCy models—spaCy but also Stanza, Flair, UDPipe
LatinCy @ 3 graphic with birthday cake
Model drop! Some (beta!) LatinCy releases ahead of Friday's dev meeting/"birthday" party, trained on same data as spaCy models…
- LatinCy Stanza huggingface.co/latincy/la_s...
- LatinCy UDPipe huggingface.co/latincy/la_u...
- LatinCy Flair huggingface.co/latincy/la_f...
#nlproc #digiclass
Comparative metrics for different LatinCy models—spaCy but also Stanza, Flair, UDPipe
LatinCy @ 3 graphic with birthday cake
Model drop! Some (beta!) LatinCy releases ahead of Friday's dev meeting/"birthday" party, trained on same data as spaCy models…
- LatinCy Stanza huggingface.co/latincy/la_s...
- LatinCy UDPipe huggingface.co/latincy/la_u...
- LatinCy Flair huggingface.co/latincy/la_f...
#nlproc #digiclass
Starting to explore different frameworks more this year as part of the next phase of LatinCy development, thanks for sharing... great to see all of the energy in this space!
tree visualisation from latinCy for a sentence from Caesar's Gallic Wars
Instead of playing video games, like an idiot I spent my weekend trying out some new-ish Latin parsers.
latinCy is pretty impressive, and does a great job for short sentences. It's also easy to integrate into a Python ecosystem and style so it looks great...
overall, the last 2-3 years have been an epochal change for automatic syntactic analysis of Latin, it is extremely exciting.
Entrevista en CLARIN a María Teresa Almeida (IATEXT) sobre su tesis sobre mujeres en florilegios latinos (s. XVI). Destaca el uso de Lemmata Embeddings for Latin, LatinAffectus (LiLa), LatinCy y Google Colab para adaptar métodos digitales a corpus filológicos. #CLARIN
www.clarin.eu/blog/tour-de...
Really looking forward to talking to the Brooklyn College community about LLM curiosity & tinkering in both research and teaching!
Prodigy interface for LatinCy NER annotations... here Ov. Met. 2.5 and a PERSON label assigned to Mulciber
Prodigy interface for LatinCy NER annotations... here Ov. Met. 2.5 and a PERSON label assigned to Mulciber; "Ask LLM" button queries an "NER Assistant" and provides more context
Super-experimental at this point—but I am now embedding a local LLM inside Prodigy to assist with NER annotations...
1. Using `correct` recipe, LatinCy model tags likely entities
2. Optional "Ask LLM" button queries Mistral based on text/tags
3. Add'l RAG runs over the project's NER guidelines
Thanks! And fwiw this is ollama+mistral7b—will eventually trial different models.
Sharing this workflow after reading the "The ultimate guide to optimizing annotation workflows" post from @inesmontani.bsky.social
cf. bsky.app/profile/ines...
As I said, super-experimental... the LLM responses are at the moment *just okay*—but often enough to save me a quick internet search or other reference check, just nice to have right there in the same interface
Prodigy interface for LatinCy NER annotations... here Ov. Met. 2.5 and a PERSON label assigned to Mulciber
Prodigy interface for LatinCy NER annotations... here Ov. Met. 2.5 and a PERSON label assigned to Mulciber; "Ask LLM" button queries an "NER Assistant" and provides more context
Super-experimental at this point—but I am now embedding a local LLM inside Prodigy to assist with NER annotations...
1. Using `correct` recipe, LatinCy model tags likely entities
2. Optional "Ask LLM" button queries Mistral based on text/tags
3. Add'l RAG runs over the project's NER guidelines
Perhaps of interest, spec. section on "Use models as annotators"... explosion.ai/blog/optimiz.... Planning on sharing some LLM-assisted annotation workflows from LatinCy NER soon as well.
Really looking forward to talking to the Brooklyn College community about LLM curiosity & tinkering in both research and teaching!
Do you have questions about AI and LLMs? Are you in the NYC area? I've asked a friend, colleague, and former student to come give all us humanist types the basics. It will be excellent! @cuny.edu @sentantiq.bsky.social @isawnyu.bsky.social
Python code for using latincy-preprocess to change u → v in Latin text, e.g. uerbum → verbum
Ever have a lot of all-u Latin text, ever need a superfast way to replace only the consonants, i.e. uerbum → verbum... new feature in latincy-preprocess v0.1... github.com/diyclassics/... #nlproc #digiclass
Python code demonstrating long-s correction using latincy-preprocess, e.g. "funt in fundamento reipublicae ftatua" → "sunt in fundamento reipublicae statua"
Ever have like a million long-s errors in your Latin OCR, ever need a superfast way to correct them against a Latin character ngram model... new feature in latincy-preprocess v0.1... github.com/diyclassics/... #nlproc #digiclass #digiclafs
Python code for using latincy-preprocess to change u → v in Latin text, e.g. uerbum → verbum
Ever have a lot of all-u Latin text, ever need a superfast way to replace only the consonants, i.e. uerbum → verbum... new feature in latincy-preprocess v0.1... github.com/diyclassics/... #nlproc #digiclass
Another #claudecode milestone... this is the first package I've put out there with a Rust implementation... cc: @rust-lang.org
Python code demonstrating long-s correction using latincy-preprocess, e.g. "funt in fundamento reipublicae ftatua" → "sunt in fundamento reipublicae statua"
Ever have like a million long-s errors in your Latin OCR, ever need a superfast way to correct them against a Latin character ngram model... new feature in latincy-preprocess v0.1... github.com/diyclassics/... #nlproc #digiclass #digiclafs