Michael Collins CSColumbia Columbia University Form

Instructions and help about Michael Collins CSColumbia Columbia University

so in last week's lectures we developed log linear models this very new way of looking at modeling for natural language processing and also parameter estimation in natural language processing in the current week's lectures we are going to look at various applications of log linear models to problems in natural language processing and the first problem we're going to cover in this segment is going to be the application of log linear models to tagging problems we'll see this is a very important example of how log linear models can be used in NLP and we'll see how to develop powerful alternatives to the hidden Markov model taggers that you saw earlier in this class so historically log linear models for tagging or I'll often also referred to as maximum entropy Markov models why is that well log linear models sometimes referred to particularly in the earlier literature as maximum entropy models so we won't really go into the reason for that naming but it's worth bearing in mind because some