|
I'm learning statistical CCG parsing by following Julia Hockenmaier's work. In her PhD thesis, section 5.3, she defined four different kinds of expansion, which made a distinction depending on the different positions of the head daughter. The way she distinguish the head and non-head is through their functional: complements, adjuncts or punctuation marks. In the generative process, the head daughter is constructed at first, then is the non-head daughter, so this distinction is important as it affects the whole structure of the parsing tree. My questions are: (1) What is the intuition of making the distinction between the head and non-head? (2) What are the advantages of this distinction comparing to treating them as conditional independent giving the parent? |
