NLP Studio for Corpora Correction and Annotation

A Java-based toolkit for correcting interlinear glossed text and generating CoNLL-U output, with both:

• a CLI pipeline for document processing,
• and a desktop application for interactive rule editing, annotation settings, workspace management, and preview.

This project was originally made for the Chuj project at Université de Montréal and is designed for low-resource language workflows based on glossed data and CoNLL-U export.

---

Overview

The repository is organized as a multi-module Maven project:

• core — correction, alignment, parsing, annotation, CLI, and CoNLL-U generation
• backend — Spring Boot services, persistence, rule/settings management
• app — JavaFX desktop wrapper embedding the frontend and backend
• frontend — Vue/Vite frontend bundled into the desktop application

---

Features

• correction of glossed interlinear entries using YAML rules
• annotation to CoNLL-U
• YAML-driven annotation configuration
• lexicon- and extractor-based annotation rules
• desktop UI for:
  • workspace entry management
  • correction preview
  • CoNLL-U preview
  • rule editing
  • annotation settings editing

---

Requirements

• Java 21
• Maven 3.9+
• For desktop packaging on Windows:
  • a JDK including jpackage
• For frontend-only development:
  • Node.js and npm

---

Repository structure

.
├── app/        # JavaFX desktop application
├── backend/    # Spring Boot backend
├── core/       # CLI + core NLP pipeline
├── docs/       # Documentation for the project
├── frontend/   # Vue/Vite frontend
└── scripts/    # packaging helpers

CLI usage

The core module provides a command-line interface.

Build

mvn -pl core -am clean package

This produces the following file :

core/target/nlp-studio-core-0.1.0-all.jar

Available commands

1. Prepare CoNLL-U from an input document

java -cp core/target/nlp-studio-core-0.1.0-all.jar org.titiplex.Main prepare input.docx correction.yaml annotation.yaml output.conllu

This command:

• reads a .docx or .txt input file,
• applies correction rules from correction.yaml,
• applies annotation settings from annotation.yaml,
• writes the resulting output.conllu.

2. Generate a corrected DOCX

java -cp core/target/nlp-studio-core-0.1.0-all.jar org.titiplex.Main correct-docx input.docx correction.yaml corrected.docx

3. Generate corpus statistics

java -cp core/target/nlp-studio-core-0.1.0-all.jar org.titiplex.Main stats input.docx correction.yaml stats.txt

Backward-compatible mode

The legacy 4-argument mode is still supported:

java -cp core/target/nlp-studio-core-0.1.0-all.jar org.titiplex.Main input.docx correction.yaml annotation.yaml output.conllu

---

Desktop application

The desktop application is a JavaFX container that starts an embedded Spring Boot backend and loads the bundled Vue frontend.

Run in development

From the repository root:

mvn -pl core,backend,app -am clean install -DskipTests
mvn -f app/pom.xml javafx:run

Frontend-only development

If you want to work on the frontend separately:

cd frontend
npm install
npm run dev

Useful frontend commands:

npm run build
npm run test
npm run typecheck

---

Packaging

Verify that the project builds correctly:

mvn clean verify

Build the desktop jar

mvn -pl app -am -Pdesktop-prod clean package

Generated file:

app/target/nlp-studio-app-0.1.0-all.jar

Create Installers

This project can build and publish installers, for each specific OS. For that, please read packaging/README.md.

---

Notes on the desktop build

The app build automatically:

• installs Node.js and npm through Maven,
• runs npm ci,
• runs the frontend build,
• copies the built frontend into the desktop application resources.

So in most cases, you do not need to build the frontend manually before packaging the desktop application.

---

Data and configuration

The pipeline relies on YAML-based resources such as:

• correction rules
• annotation definitions
• POS and feature definitions
• lexicons
• extractors
• gloss mapping

This makes the system extensible and suitable for iterative linguistic work without hardcoding every rule in Java.

---

Testing

Run all tests:

mvn test
# or
mvn clean verify "-Dskip.frontend=true"
# then
cd frontend
npm ci
npm run typecheck
npm run playwright:install
npm run test
npm run test:e2e
npm run test:coverage
npm run build

Run the full project build:

mvn clean package

---

Status

This branch focuses on an integrated NLP studio workflow rather than only a standalone converter:

• CLI processing remains available
• desktop editing and preview are first-class
• backend-managed rules and annotation settings are part of the current architecture

---

Documentation

The documentation is built using MkDocs and is available at this repository's GitHub Pages.

To preview the doc and edit it live:

mkdocs serve --livereload

To build the doc (generates files in site/) :

mkdocs build

To publish the doc in your repository under github pages, in the gh-pages branch :

mkdocs gh-deploy

---

License

This projects runs under the GPL-v3 license, please see LICENSE