OpenNLP

For JVM developers who want classic text processing without a Python sidecar, the Apache Software Foundation's OpenNLP library covers the standard pipeline: sentence detection, tokenizing, part-of-speech tags, lemmas, named entities, chunking, parsing, language identification, and document categorization. Components use built-in statistical learners, maximum entropy by default, plus perceptron, naive Bayes, and SVM, and each can be retrained on your own annotated corpus. A newer opennlp-dl module loads ONNX transformer models through ONNX Runtime, with a GPU variant, so BERT-style models run behind the same API. Use it through the Java API or a CLI, including inside Spark, Flink, or NiFi jobs; UIMA annotators, Morfologik stemming, and spell checking ship as extra modules. The 3.x line, still in milestone releases, splits the toolkit into smaller Maven artifacts, makes the core ME classes thread safe, and needs JDK 21, while the Java 17 based 2.x branch still gets fixes. Demo models exist for many languages, but the project advises training your own for production. Apache 2.0 allows unrestricted commercial use.