When looking into NLTK to support a Twitter classifier, even though I had no NLP experience, the NLTK syntax/interface was sufficiently self-contained.

I was mostly interested in tagging and chunking and found it easy to reason about and use the relevant NLTK features with little auxiliary reading.

The only issues I found were with NLTK itself, e.g., Python version support, loading time, packages.