LIMA — Libre Multilingual Analyzer¶
LIMA is a multilingual natural language processing (NLP) analyzer developed by CEA LIST, in its Text and Image Semantic Analysis laboratory (LASTI). It is free software, released under the MIT license.
LIMA combines two complementary approaches:
- Neural modules (built on libtorch, the PyTorch C++ library) for tokenization, part-of-speech tagging and morphological features, lemmatization and dependency parsing, with models for more than 60 languages trained on Universal Dependencies treebanks.
- ModEx rules: a powerful finite-state rule formalism to extract entities, relations and events in domains where no annotated data exists.
Results are produced in CoNLL-U by default.
-
:material-rocket-launch:{ .lg .middle } Getting started
Install LIMA and analyze your first text in a few minutes.
-
:material-download:{ .lg .middle } Install
Python package, Docker image, Linux packages or build from source.
-
:material-console:{ .lg .middle } Using LIMA
Models, command line, Python API, pipelines and configuration.
-
:material-book-open-variant:{ .lg .middle } Reference
Configuration files, ModEx rules, process units, Python and C++ APIs.
Features¶
- Fast C++ engine with a native Python API and a simple graphical interface.
- Neural Universal Dependencies analysis: tokenization, sentence splitting, UPOS and morphological features, lemmatization and labeled dependency parsing.
- Named entity recognition for English and French.
- Rule-based information extraction with ModEx: extract new kinds of entities, relations or events without annotated data.
- Legacy linguistic pipeline for English and French: full-form dictionaries, hyphenated and concatenated words splitting, idiomatic expressions, rule-based parsing, coreference resolution and semantic analysis.
- A plugin architecture: pipelines are assembled from configuration files and can be extended without recompiling LIMA.
- Regression testing and evaluation tools.
Get in touch¶
- Questions, bug reports and suggestions: open a GitHub issue.
- Want to help? Read the contributing guide.
- To integrate LIMA in your projects or to develop resources and models for your needs, contact CEA LIST.