Hybrid approaches for NLP applications

Hybrid approaches sometimes really help us improve the result, of our NLP applications. For example, if we are developing a grammar correction system, a module that identifies multiword expressions such as kick the bucket, and a rule-based module that identifies the wrong pattern and generates the right pattern. This is one kind of hybrid approach. Let's take a second example for the same NLP application. You are making a classifier that identifies the correct articles (determiners - a, an, and the) for the noun phrase in a sentence. In this system, you can take two categories - a/an and the. We need to develop a classifier that will generate the determiner category, either a/an or the. Once we generate the articles for the noun phrase, we can apply a rule-based system that further decides the actual determiner for the first category...