i202 Fall 2010 School of Information, UC Berkeley

 In today's lecture, Bob mentioned the differences between traditional linguistic and statistical approaches to language processing.  In my undergraduate linguistics major, I studied both syntax and "grammar engineering," for which we had to construct representations like the one below.  If anyone ever had to diagram sentences in middle or high school, it's kind of like that on steroids: 

For my class, we built a small rule-based English grammar to generate such structures.  It was a great way to understand hands on how complex syntactic structures are, and how much ambiguity there is in natural language, which are two of the main reasons that pure linguistic approaches to language processing are so computationally intensive, and why therefore in practice probabilistic models are much more common.