Messages
We submitted exam results to the study administration earlier today, so these will hopefully become visible for students sometime today or tomorrow. Lilja ?vrelid was the second examiner, and we were both quite impressed with the overall level of knowledge documented in the exams. The distribution of grades, accordingly, is somewhat tilted: there were two As, four Bs, 2 Cs, 1 D, and 3 Es. In case you would like to know more about how we arrived at your individual grade, I will be very open to discuss (or review) exam results with you individually; please make contact in email, if you want to get in touch for this in early January. Finally, in case you have not submitted your on-line course evaluation, please do so now. For our own benefits (and that of future student generations), we very much depend on your critical feedback. God Jul!
The model solution to our final assignment is now on-line; we will review key points of our implementation in the last laboratory session for this fall on Thursday this week (at the usual time and place). Please note that we do not provide a solution to the 'bonus material' of this problem set, i.e. question (4); accordingly, the integration of our statistical parser and PoS tagger is not directly exam-relevant. Speaking of the exam, please observe that utility materials (e.g. our text book or slides copies) are not allowed.
As of today, the remainder of problem set (2b) is (finally) available. Please note that I found a few (relatively minor) deficiencies in my pre-processing of the Penn Treebank, hence I have also provided updates to the files wsj.mrg and test.mrg in the IFI Linux universe; to get access to these updates, I recommend you delete the two files and question, log out and back in again (which should give you fresh copies of these files). The correct size of wsj.mrg should be 18243968 bytes.
Due to popular request, we offer an additional laboratory session this week: tomorrow (Thursday, November 11), between 16:15 and 18:00, Johan will be available in our usual laboratory, C207 in Vilhelm Bjerknes hus. Please prepare carefully for our laboratory time, i.e. try to work through as much of the current problem set as is humanly possible. On the basis of your own work, bring along to the laboratory specific problems and questions.
Finally, the complete problem set (2a) is available. Note that, as discussed in class today, we have scaled down our ambitions somewhat and leave the adaptation of the generalized chart parser to deal with (large) probabilistic context-free grammars to the final problem set (2b). Also, observe that the slide copies posted to the course page earlier today were incomplete; I have now posted a corrected version, so in case you accessed the file earlier today, please make sure to get the latest revision.
Note that we opted to push back the submission deadlines for the remaining two problem sets (2a) and (2b) by one week each; this is really the latest submission dates we can do, i.e. there is no further room for postponing things this fall.
My apologies: I have yet to finalize assignment (2a) and post it to the course web site. Our next problem set should became available no later than tomorrow afternoon (after the lecture); watch this space for updates!
As we discussed in class yesterday, we want to avoid the collision of our laboratory tomorrow with dagen@ifi. Thus, there is no INF4820 laboratory tomorrow, and instead we will look for a suitable time next week to schedule an extra laboratory.
For those of you who started their MSc programme this fall and have yet to decide on the direction to explore in your final MSc project. The IFI group in Language Technology has a number of MSc projects available. Besides the project ideas posted in the central database, we are also very open to discuss with you individually and jointly define a project that matches your interests, background, and the research expertise at the group. Please just make contact with Stephan, Lilja, or Jan Tore.
I had posted a model solution for the HMM assignment early today; in the laboratory, we jointly found a number of ways of improving this work. Also, I was embarassed that the file was not anally commented (as Johan clearly expects of your submissions). Hence, I just posted a revised and extended model solution, which I would ask everyone to retrieve and study carefully. We mean it: HMMs and Viterbi are highly relevant for this class, including the final exam. Also just now, I posted more of the weekly plan for the remainder of the semester. Specifically, please note the required reading for the lecture next week, where we transition to a new (and somewhat more complex) search problem: parsing with probabilistic context-free grammars.
Talking to a colleague just now, I realize there are another two useful Lisp references, both available on-line for quick look-up of functions or various language concepts more generally: Common Lisp the Language (Second Edition) by Guy L. Steele Jr. strikes a good balance of language reference and (subtle) humor; the Common Lisp HyperSpec is less suited for reading, but much more densely linked.
Last night, Johan emailed everyone with comments on your submissions to assignment (1b). There is substantial variation in scores this time: ranging from 21 to 92 points. Accordingly, some people have accumulated less than 100 points from (1a) and (1b) this far, which is below the 50% minimum threshold. Don't panic! Recall that you are required to amass at least 150 points for the first obligatory exercise, i.e. assignments (1a) through (1c). The submission deadline for (1c) is more than a week away, and we will have two laboratories (today and Thursday next week) to work jointly with you on this assignment. The key is to start now and aim to get all the way through before next Thursday; this way, you can take advantage of Johan and me being available for consultation. Also, remember that we are happy to receive queries in email, in case you get stuck somewhere.
Our very polished model solution is now available; as all fine Lisp code, ours is generously commented and a pleasure to read. Please take a closer look at our code prior to the laboratory tomorrow and form an opinion about our approach to the problem. Also, as of today, the last part of the first obligatory exercise is available, i.e. Assignment (1c). For the next (almost) three weeks, we will be working on the implementation of Hidden Markov Models (HMMs). Finally, Johan sent out feedback to your submissions for assignment (1a) on Monday this week; in case you have submitted something for the first assignment but have not yet heard back from Johan, please let us know immediately.
We just published our model solution for problem set (1a). Please study our approach to solving the various parts of this warm-up assignment before the laboratory tomorrow. We will not go through the entire model solution in the laboratory, but rather seek to highlight important aspects; we would be very happy to discuss alternate solutions, i.e. in case you took a different approach, please bring along an opinion tomorrow on which solution you think is more appropriate. Also, please note that our laboratory sessions are moved to the room next door: C207 in Vilhelm Bjerknes Hus.
I stayed home not perfectly well today, hence did not get to releasing our next problem set before now. In the laboratory tomorrow, I will lecture for twenty or so minutes on the use of Lisp structures and the integrated development environment. Even though the submission deadline for problem set (1a) is not until next Monday, we would like to encourage people to start on (1b) as soon as possible. Ideally, you would use tonight to finish up (1a) or reveal insurmountable problems that require our help, and then show up for the laboratory in the morning, bright awake and cheerful, and start jointly with us on problem set (1b).
We are continually adding content to the course web pages for this semester; please check this space frequently. The first meeting of the class is tomorrow (August 31) at 12:15 in Seminar Room 3B at the Department of Informatics.
This course in 2010 will be quite similar to the successful versions of 2009 and 2010. Specifically, we will assume the same NLP text book (Jurafsky & Martin, 2008), as background to various search problems to be studied and implemented; please obtain a copy of this book for the start of the semester. Additionally, we will read parts of Seibel (2005), as a general introduction to Common Lisp and intelligent problem solving; this book is available on-line, hence no need to purchase a personal copy (unless you decide you wanted to :-). Please watch this space for updates over the next few days. Remember that attendance of the first lecture, on August 31, is obligatory.