---
language:
  - la
license: mit
tags:
  - udpipe
  - latin
  - nlp
  - pos
  - lemmatization
  - dependency-parsing
  - universal-dependencies
  - digital-humanities
  - classics
  - latincy
library_name: udpipe
model-index:
  - name: la_udpipe_latincy
    results:
      - task:
          type: token-classification
          name: Part-of-Speech Tagging
        dataset:
          type: universal_dependencies
          name: UD Latin (6 treebanks)
        metrics:
          - type: f1
            value: 94.07
            name: UPOS F1
      - task:
          type: token-classification
          name: Morphological Features
        dataset:
          type: universal_dependencies
          name: UD Latin (6 treebanks)
        metrics:
          - type: f1
            value: 80.82
            name: UFeats F1
      - task:
          type: token-classification
          name: Lemmatization
        dataset:
          type: universal_dependencies
          name: UD Latin (6 treebanks)
        metrics:
          - type: f1
            value: 92.99
            name: Lemma F1
      - task:
          type: structured-prediction
          name: Dependency Parsing
        dataset:
          type: universal_dependencies
          name: UD Latin (6 treebanks)
        metrics:
          - type: f1
            value: 76.48
            name: UAS F1
          - type: f1
            value: 71.57
            name: LAS F1
---

# LatinCy UDPipe (la_udpipe_latincy_multi)

A UDPipe 1 model for Latin trained on six harmonized Universal Dependencies treebanks from [LatinCy](https://huggingface.co/latincy). Provides tokenization, POS tagging, lemmatization, morphological features, and dependency parsing in a single file.

## Highlights

- **Single-file model** (72.3 MB) -- works offline, no GPU needed
- **6 treebanks**: ITTB, LLCT, PROIEL, Perseus, UDante, CIRCSE (48,754 training sentences)
- **Drop-in replacement** for existing UDPipe Latin workflows in R, Python, CLI, Java, C#, and Perl
- Optimized with [pre-trained word embeddings](https://huggingface.co/latincy/latincy-w2v), swap parser, and two-model tagger

## Quick Start (R)

The primary audience for this model is R users working with the [`udpipe`](https://CRAN.R-project.org/package=udpipe) package.

```r
library(udpipe)

# Download the model (one time)
model_url <- "https://huggingface.co/latincy/la_udpipe_latincy/resolve/main/la_udpipe_latincy_multi.udpipe"
model_path <- "la_udpipe_latincy_multi.udpipe"
if (!file.exists(model_path)) {
  download.file(model_url, model_path, mode = "wb")
}

# Load and annotate
model <- udpipe_load_model(model_path)
text <- "Gallia est omnis divisa in partes tres."
result <- udpipe_annotate(model, x = text)
df <- as.data.frame(result)
print(df[, c("token", "upos", "lemma", "dep_rel")])
```

Output:

```
    token  upos  lemma dep_rel
1  Gallia PROPN Gallia   nsubj
2     est   AUX    sum     cop
3   omnis   DET  omnis     det
4  divisa   ADJ divisa    root
5      in   ADP     in    case
6  partes  NOUN   pars     obl
7    tres   NUM   tres  nummod
8       .  PUNCT     .   punct
```

## Quick Start (Python)

```python
from ufal.udpipe import Model, Pipeline

model = Model.load("la_udpipe_latincy_multi.udpipe")
pipeline = Pipeline(model, "tokenize", Pipeline.DEFAULT, Pipeline.DEFAULT, "conllu")
result = pipeline.process("Gallia est omnis divisa in partes tres.")
print(result)
```

Install the Python bindings with `pip install ufal.udpipe`.

## Quick Start (CLI)

```bash
# Download the model
curl -L -o la_udpipe_latincy_multi.udpipe \
  https://huggingface.co/latincy/la_udpipe_latincy/resolve/main/la_udpipe_latincy_multi.udpipe

# Annotate text
echo "Gallia est omnis divisa in partes tres." | \
  udpipe --tokenize --tag --parse la_udpipe_latincy_multi.udpipe
```

## Model Description

| Property | Value |
|----------|-------|
| **Author** | Patrick J. Burns / [LatinCy](https://huggingface.co/latincy) |
| **Model type** | UDPipe 1 ([MorphoDiTa](https://ufal.mff.cuni.cz/morphodita) tagger + [Parsito](https://ufal.mff.cuni.cz/parsito) parser) |
| **Language** | Latin |
| **License** | MIT |
| **File size** | 72.3 MB (larger than typical single-treebank UDPipe models due to multi-treebank training data and embedded word vectors) |
| **Version** | 0.2 |
| **Framework** | UDPipe 1 |

## Training Data

Trained on harmonized data from 6 Universal Dependencies Latin treebanks, prepared by the [LatinCy](https://huggingface.co/latincy) treebank pipeline. Treebanks are harmonized to consistent annotation standards before combining.

| Treebank | Full Name | Domain |
|----------|-----------|--------|
| [**ITTB**](https://universaldependencies.org/treebanks/la_ittb/) | Index Thomisticus Treebank | Scholastic Latin (Thomas Aquinas) |
| [**LLCT**](https://universaldependencies.org/treebanks/la_llct/) | Late Latin Charter Treebank | Medieval legal charters |
| [**PROIEL**](https://universaldependencies.org/treebanks/la_proiel/) | PROIEL Treebank | Vulgate Bible, historical texts |
| [**Perseus**](https://universaldependencies.org/treebanks/la_perseus/) | PerseusDep Latin Treebank | Classical Latin (Caesar, Cicero, etc.) |
| [**UDante**](https://universaldependencies.org/treebanks/la_udante/) | UDante Treebank | Dante Alighieri (De vulgari eloquentia, etc.) |
| [**CIRCSE**](https://universaldependencies.org/treebanks/la_circse/) | CIRCSE Latin Treebank | LASLA-derived classical texts |

**Split sizes:** 48,754 train / 4,613 dev / 6,423 test sentences.

## Training Procedure

Trained using `ufal.udpipe` Python bindings (UDPipe 1, morphodita_parsito backend).

**Tokenizer:**
- 100 epochs max, batch size 50, learning rate 0.005, dropout 0.1
- Early stopping enabled (stopped at epoch 42)

**Tagger** (two-model MorphoDiTa configuration):
- Model 1: UPOS, XPOS, morphological features (20 iterations, tagger templates)
- Model 2: Lemma (20 iterations, lemmatizer templates)

**Parser** (Parsito):
- Swap transition system (handles non-projective dependencies)
- Static lazy oracle, 200-unit hidden layer
- [Pre-trained word2vec embeddings](https://huggingface.co/latincy/latincy-w2v) (CBOW-300, 142K vocab, 96.1% form coverage)
- 10 iterations max, early stopping (stopped at iteration 9)

## Evaluation Results

Evaluated on held-out test data (6,423 sentences) with gold tokenization using `conll18_ud_eval.py`.

### Overall Scores (v0.2)

Weighted average across 6 test treebanks (106,489 words).

| Metric | F1 |
|--------|-----|
| UPOS | 94.07 |
| UFeats | 80.82 |
| Lemma | 92.99 |
| UAS | 76.48 |
| LAS | 71.57 |
| CLAS | 67.67 |
| MLAS | 53.86 |
| BLEX | 64.44 |

### Per-Treebank Breakdown (v0.2)

| Treebank | UPOS | UFeats | Lemma | UAS | LAS |
|----------|------|--------|-------|-----|-----|
| ITTB | 97.03 | 88.36 | 97.90 | 83.69 | 80.33 |
| LLCT | 97.96 | 87.15 | 96.90 | 91.62 | 89.91 |
| PROIEL | 94.73 | 76.26 | 94.44 | 74.69 | 69.44 |
| Perseus | 90.26 | 70.43 | 85.98 | 66.13 | 57.73 |
| UDante | 87.82 | 75.71 | 85.19 | 64.02 | 55.80 |
| CIRCSE | 88.48 | 70.28 | 86.16 | 53.52 | 45.64 |

### Comparison with Stock UDPipe UD 2.5 Models

The [UDPipe UD 2.5 models](https://lindat.mff.cuni.cz/repository/xmlui/handle/11234/1-3131) (Straka & Straková, December 2019) are single-treebank models trained with default hyperparameters and distributed with UDPipe. Three exist for Latin: ITTB, Perseus, and PROIEL. No stock models exist for LLCT, CIRCSE, or UDante.

This model differs from stock models in three ways: (1) training on all 6 treebanks rather than one, (2) harmonized annotations across treebanks, and (3) optimized hyperparameters including pre-trained embeddings and a swap parser.

**LAS F1 comparison (v0.2):**

| Test Set | LatinCy | Stock-ITTB | Stock-Perseus | Stock-PROIEL |
|----------|---------|-----------|--------------|-------------|
| ITTB | **80.33** | 66.94 | 36.75 | 39.31 |
| LLCT | **89.91** | 31.20 | 23.19 | 29.35 |
| PROIEL | **69.44** | 44.79 | 39.14 | 50.67 |
| Perseus | **57.73** | 39.58 | 42.26 | 34.49 |
| UDante | **55.80** | 42.47 | 26.63 | 30.69 |
| CIRCSE | **45.64** | 26.81 | 25.77 | 27.43 |

**Observations:**
- The LatinCy model outperforms stock models across all treebanks, including on each stock model's own training domain (ITTB +13.39, Perseus +15.47, PROIEL +18.77 over best stock)
- Single-treebank stock models show expected domain sensitivity when applied cross-domain (e.g., Stock-ITTB on LLCT: 31.20 LAS)
- Multi-treebank training on harmonized data provides broader coverage across Latin text types

### Cross-Framework Comparison (LatinCy v3.9)

All models trained on the same harmonized treebank data. Scores on held-out test sets unless noted. NER scores are on dev (no test set exists).

| Metric | LatinCy<br>UDPipe 0.2 | LatinCy<br>Stanza 0.3 | LatinCy<br>Flair 0.3 | LatinCy<br>spaCy trf 3.9 |
|--------|------------|---------------------|---------------------|----------------|
| UPOS | 94.07 | 97.65 | **98.02** | 97.34 |
| UFeats | 80.82 | 93.93 | -- | **93.95** |
| Lemma | 92.99 | **97.87** | 97.41 | 94.63 |
| UAS | 76.48 | **86.95** | -- | 86.91 |
| LAS | 71.57 | **83.23** | -- | 82.04 |
| NER F1 | -- | 90.22 | **92.22** | 91.14 |

UDPipe's strength is portability: a single file usable from R, Python, CLI, Java, C#, and Perl with no GPU and no framework dependencies. Stanza leads on lemma and dependency parsing. Flair 0.3 (Latin BERT) leads on UPOS and NER.

## R Usage Guide

### Installation

```r
install.packages("udpipe")
```

### Download and Load

```r
library(udpipe)

# Download the model
model_url <- "https://huggingface.co/latincy/la_udpipe_latincy/resolve/main/la_udpipe_latincy_multi.udpipe"
model_path <- "la_udpipe_latincy_multi.udpipe"
if (!file.exists(model_path)) {
  download.file(model_url, model_path, mode = "wb")
}

# Load
model <- udpipe_load_model(model_path)
```

### Basic Annotation

```r
text <- "Arma virumque cano, Troiae qui primus ab oris
Italiam fato profugus Laviniaque venit litora."
result <- udpipe_annotate(model, x = text)
df <- as.data.frame(result)
```

### Tidyverse Integration

```r
library(dplyr)

# Count POS tags
df %>%
  count(upos, sort = TRUE)

# Extract nouns with their lemmas
df %>%
  filter(upos == "NOUN") %>%
  select(token, lemma, feats)

# Get dependency relations
df %>%
  select(token_id, token, head_token_id, dep_rel, upos)
```

### Replacing a Stock Model

If you previously used a stock UDPipe Latin model, replace it by pointing to this model file instead:

```r
# Before (stock model)
# model <- udpipe_download_model(language = "latin-ittb")
# model <- udpipe_load_model(model$file_model)

# After (LatinCy model)
model <- udpipe_load_model("la_udpipe_latincy_multi.udpipe")
# Everything else stays the same
```

## Limitations

- **UFeats trade-off:** v0.2 improves UPOS (+0.8) and LAS (+0.3) over v0.1 but trades ~1.7 UFeats points. The tagger template configuration prioritizes tagging and parsing accuracy.
- **UDPipe 1 architecture:** Feature-based model (no transformers). For higher accuracy, consider [LatinCy spaCy models](https://huggingface.co/latincy) (`la_core_web_trf`: 79.8 LAS).
- **Gold tokenization evaluation:** Reported scores use gold tokenization. Real-world performance on raw text depends on the tokenizer (99.6% F1 on held-out data).

## References

- Straka, M., and Straková, J. 2017. "Tokenizing, POS Tagging, Lemmatizing and Parsing UD 2.0 with UDPipe." In Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. Vancouver, Canada: Association for Computational Linguistics. 88–99. https://aclanthology.org/K17-3009/.
- Straka, M., Hajič, J., and Straková, J. 2016. "UDPipe: Trainable Pipeline for Processing CoNLL-U Files Performing Tokenization, Morphological Analysis, POS Tagging and Parsing." In Calzolari, N., Choukri, K., Declerck, T., Goggi, S., Grobelnik, M., Maegaard, B., Mariani, J., et al. eds. Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16). Portorož, Slovenia: European Language Resources Association (ELRA). 4290–97. https://aclanthology.org/L16-1680/.

## Citation

```bibtex
@misc{burns2026latincyudpipe,
  author = {Burns, Patrick J.},
  title = {{LatinCy UDPipe (la\_udpipe\_latincy\_multi)}},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/latincy/la_udpipe_latincy},
}
```

## Acknowledgments

This work was supported in part through the [NYU IT High Performance Computing](https://sites.google.com/nyu.edu/nyu-hpc/about/acknowledgement-statement) resources, services, and staff expertise.