Kathleen Mckeown

There is absolutely no reason you should take this course unless you want to have the most miserable semester of your life.

A very good class and professor. Taking Artificial Intelligence with Professor Mckeown convinced me to pursue AI and computer science further. Very highly recommended.

Both the professor and the class materials are awesome. This is one of the best courses I have taken in Columbia. Professor McKeown’s class provides a very comprehensive overview of NLP. Topics include language models, neural networks, word embeddings, sentimental analysis, POS tagging, parsing, semantics, machine translation, summarization. The class materials and the homework assignments are a very good combination of theory and coding. For theory, we learn parsing algorithms, read word embedding papers, study the math formula behind those frequently used neural networks. For practice, we implement our own model to do stance prediction, sentimental analysis and machine translation. Assignments often go beyond simple code implementation. Instead, most of the assignment focuses on analysis of the models that we built – analysis of model performance, error analysis on frequent mistakes made by the model, comparison of different methods and models. These analysis techniques turn out to be very useful in real-word applications, and really distinguish those who know NLP from those who only know how to use packages. Professor McKeown is a very nice person who actually cares about every student that she teaches. She is the first professor that I’ve met with in Columbia who does her best to really remember almost everyone’s name in the classroom (it’s a class of approximately 100 stu…

I took this course in the Spring of 2004 and found it challenging and useful. The course covered the requisite material thoroughly, and offered intriguing (and sometimes quite difficult) programming assignments. Other reviewers complain about the focus on 'boring algorithmic parts' and lack of attention to philosophical implications. This is both ironic and inaccurate on a number of counts. First, the early part of the course was dedicated to a thorough consideration of the difference between human-like and rational behavior; in addition, a number of readings were posted over the term that addressed just these issues. Second, this course is offered in a computer science department; the capacity of computer scientists to have impact on scientific or philosophical problems is founded in the algorithmic details. Glossing over this material would cheat serious students who are willing to do the work. I enjoyed the course and still discuss and apply the material I learned.

Horrible Horrible Class. Homework is almost completely unrelated to what is gone over in class. Class is boring and long with PollEverywhere scattered here and there. Homework basically consists of throwing a bunch of model at a wall and see what sticks, can be fast if you have some applied machine learning background but you literally learn nothing from the homework other than getting frustrated with the libraries. Only upside is Kathy is a great human being who genuinely cares about her students - she literally remembers everyone's name, but that's about it. She doesn't make the course any more interesting.

Prof McKeown is a nice lady. Let's say that first. She's a pretty lousy professor, though. It's hard to tell whether she's just bored with the material, or doesn't know how to make it interesting, or just doesn't care. In any event, her classes consist of a series of slides that are pretty much straight from the textbook, and a lecture delivered in monotone. She seems actively disinterested in any of the interesting intellectual, scientific, or philosophical implications of the material, and instead grinds through the most boring algorithmic parts, yet without enough detail or mathematical sophistication to make it worthwhile. Avoid if possible.

She did a good job of covering the material and explaining what she expected for the midterms. About what you'd expect for this level class. There was a curve. http://www1.cs.columbia.edu/~kathy/NLP/

A very nice professor who cares enough to learn everyones name in the relatively large class. But I don't think she was born to teach. She is very well prepared in that she has everything on powerpoint, but she's not so great at answering questions that aren't in the powerpoint plan. She's also rather bad at coming up with good test questions, asking students to go through every step of an algorithm that is easy for computers, but painfully tedius for humans, and making that worth almost half the midterm. (And giving a full blue book for it). The subject is fascinating, and if no one else is teaching it, I still highly recommend taking it. Kathleen Mckeown is a great professor if you want a friend and a nice office hours buddy, is knowledgeable and cares, but just isn't a born teacher.

4 Homeworks that combine theory with practice 1 Midterm of 75 minutes 1 Final of 75 minutes

1 midterm, 1 final (both reasonable), and a number of programming assignments that you can spend as much time on as you want to.