A core-tools statistical NLP course
Dan Klein · 2005
In the fall term of 2004, I taught a new statistical NLP course focusing on core tools and machine-learning algorithms. The course work was organized around four substantial programming assignments in which the students implemented the important parts of several core tools, including language models (for speech reranking), a maximum entropy classifier, a part-of-speech tagger, a PCFG parser, and a word-alignment system. Using provided scaffolding, students built realistic tools with nearly state-of-the-art performance in most cases. This paper briefly outlines the coverage of the course, the scope of the assignments, and some of the lessons learned in teaching the course in this way.