Stanford CoreNLP

For classic linguistic annotation in Java, this Stanford NLP Group suite runs raw text through a configurable chain of annotators: tokenizing and splitting sentences, tagging parts of speech, lemmatizing, finding named entities, resolving coreference, building constituency and dependency parses, scoring sentiment, and normalizing dates and quantities with SUTime. Beyond English, model jars exist for Spanish, Hungarian, Italian, French, Arabic, German and Chinese, also mirrored on Hugging Face. You can call it from Java code, a command-line pipeline, or a bundled HTTP server on port 9000 that answers in JSON, which is how Python users usually reach it through Stanza. Release 4.6.0 (September 2026) moved the baseline to Java 11 and closed several deserialization weaknesses. Memory scales with the annotators chosen: tokenizing alone needs very little, but coreference plus parsing on long documents calls for 4 GB of heap or more. Licensing is GPL v3 or later, so shipping it inside closed software requires a commercial license from Stanford. Researchers and teams needing transparent, non-LLM text analysis get the most from it.