Tomorrow (Thursday), Andrey and Stephan will hold the final laboratory session for this semester. We have just published our own attempt at neural negation resolution (in the Microsoft GitHub repository for the course at UiO) and will use the session tomorrow to (a) admire our own code and share practical tips for hyper-parameter grid search on Abel; (b) invite everyone to critically review our code and in particular look for wasteful operations; (c) share some high-level comments on your submissions; and (d) discuss with you (which is not to say answer) any questions you might have about the exam.
We now publish the list of submitted systems which were the best in the obligatory assignment 3 (Sentiment analysis with CNNs), when evaluated on the held-out test set. Congratulations to the winners! The test set is now available in the same location as the train and dev sets (/projects/nlpl/teaching/uio/inf5820/2018/SST/ on Abel)....
We have just published the final assignment for this course, which has its submission deadline on Friday, November 16. This assignment builds on a fairly popular standard task and data set, providing more structural complexity than we have encountered so far. The first part of the assignment is intended to explore the data and enable you to navigate its structure comfortably; please start working on this part of the assignment already this week! To make available time for this preparatory work, all of us traveled today to the EMNLP 2018 conference, which means there will be no laboratory session this Thursday.
Next Tuesday, we will related the theory of gated recurrent neural networks to various practical language processing tasks, including the one of the final assignment.