---
tags:
- spacy
- latin
- sentence-segmentation
language:
- la
license: mit
model-index:
- name: la_senter
  results:
  - task:
      name: SENTS
      type: token-classification
    metrics:
    - name: Sentences F-Score
      type: f_score
      value: 0.9971
    - name: Sentences Precision
      type: precision
      value: 0.9966
    - name: Sentences Recall
      type: recall
      value: 0.9976
---

# la_senter

Latin sentence segmentation model for [LatinCy](https://huggingface.co/latincy).

| Feature | Description |
| --- | --- |
| **Name** | `la_senter` |
| **Version** | `3.9.2` |
| **spaCy** | `>=3.8.0,<3.9.0` |
| **Default Pipeline** | `senter`, `token_fix` |
| **Components** | `senter`, `token_fix` |
| **Sources** | UD_Latin-Perseus, UD_Latin-PROIEL, UD_Latin-ITTB, UD_Latin-LLCT, UD_Latin-UDante |
| **License** | `MIT` |
| **Author** | [Patrick J. Burns](https://diyclassics.github.io/) |

## Install

```bash
pip install https://huggingface.co/latincy/la_senter/resolve/main/la_senter-3.9.2-py3-none-any.whl
```

## Usage

```python
import spacy

nlp = spacy.load("la_senter")

doc = nlp("Gallia est omnis divisa in partes tres. Quarum unam incolunt Belgae.")
for sent in doc.sents:
    print(sent.text)
# Gallia est omnis divisa in partes tres.
# Quarum unam incolunt Belgae.

doc = nlp("Iphicles, frater Herculis, magna voce exclamavit; sed Hercules ipse, fortissimus puer, haudquaquam territus est.")
for sent in doc.sents:
    print(sent.text)
# Iphicles, frater Herculis, magna voce exclamavit;
# sed Hercules ipse, fortissimus puer, haudquaquam territus est.
```

## What's new in 3.9.2

This release adds **`token_fix`**, a rule-based component that runs after
`senter` to repair sentence boundaries around punctuation that the statistical
model mis-splits. It is bundled with the model and registered automatically as
a spaCy factory, so `spacy.load("la_senter")` activates it with no extra imports.

- **Parentheticals** — a `!` or `?` inside `(...)` or `[...]` no longer ends the
  sentence: `Verba (mirabile dictu!) sunt.` stays one sentence.
- **Dash asides** — the same for em/en/hyphen-delimited asides:
  `Nostraque — me miseram! — timui.` stays one sentence.
- **Orphaned closing quotes** — a sentence-final closing quote is kept with its
  sentence rather than split onto the next: `"Ita est." "Dixit."` breaks after
  the first closing quote, not before it.

## 3.9.1

Internal development version — first bundled the fixers but was never published
to HuggingFace; superseded by 3.9.2 (which corrects the component registration
and aligns the version with the `latincy-pipelines` releases).

## What's new in 3.9.0

- Case-insensitive segmentation: correctly handles lowercased and mixed-case input
- Sentence splitting on semicolons and colons
- Bracketed reference handling: `[2] O tempora...` is treated as one sentence

## Accuracy

| Type | Score |
| --- | --- |
| `SENTS_F` | 99.71 |
| `SENTS_P` | 99.66 |
| `SENTS_R` | 99.76 |

Evaluated on held-out test split from the combined UD treebanks.

## Intended use

Sentence segmentation of well-punctuated Latin text from digital editions, corpora, and scholarly sources. Not designed for punctuation-free text (scriptura continua).

## Training

Trained on five Universal Dependencies Latin treebanks using spaCy's `senter` component.