In NLP, rule-based/linguistic approaches are usually said to provide relatively high precision but low recall compared to for example machine learning approaches. I think this is often true for almost any classification task in NLP. However, I couldn't find a paper which specifically examines this phenomena. Can someone please name one? Alternatively, I'd be interested in a paper which generally compares different approaches in NLP classification tasks.
I did quite some googling but all I found is a very vague statement in here: http://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-15-285