How to combine the strengths of PCFG (sentence structure) and n-gram models (lexical co-occurrence)?
Have a look at Dan Klein's paper.