19 March 2013

One day of an NLP student’s life

Are you currently looking for a master’s program in Natural Language Processing? Feel free to consider Europe then. Yes, we all know that the US is known to be the best place to study NLP, but in case you don’t have a spare 100 000$, there is quite a bunch of European programs, including those in France and Germany for example, where you don’t need to pay a more than 500 euro (!) per year. And this also covers a medical insurance.
As you might have already understood I am a NLP student. After I have finished my Bachelor’s degree in French/German linguistics, I was looking for an interesting graduate program in the field of computational linguistics, and there was nothing like this I could find in my home country, Russia.
So, I decided to go somewhere else, and even far far away, if needed. For several quite obvious reasons I chose European education.  My choice in Europe was France and I never regretted it. Why France? First, there is a number of quite different and varied programs proposed in France, varying from NLP applied in foreign language learning to statistical NLP for different purposes. Then, France is a fantastically beautiful country. Being a student in France you can freely travel to within Europe.  And finally, I am fluent in French, and this a big advantage, because the first year of studies was conducted in French.  
I was accepted to 8 universities all over France, and chose the one in Nancy, region of Lorraine. The master’s program proposed here is called “MSc in Cognitive Science” with NLP as one of specializations. During the first year we studied a wide range of courses, like Algorithms for Artificial Intelligence, Neural networks, Logics, as well as some introductory disciplines for NLP (maths for Computer Sciences, Python programming, etc.)  A multidisciplinary program like this allows you not only to discover the huge field of Cognitive Science, but also to see the applications of NLP in interesting context. For example, my first year’s course work was about the application of NLP (language generation) in Serious Games (in virtual environments, like SecondLife). 
So, if you are curious to know, how we spend our study time on the way to the Master’s degree you are welcome to read further!
8.00-10.00 Yep, it’s hard to wake up at this hour, but I’m motivated – today we have a class of Statistical NLP, a necessary and even obligatory discipline for almost any computational linguist these days. Today we are listening to a lecture on smoothing technics for language models used in many NLP applications, like speech recognition or statistical machine translation. As usual, at the end of the class the professor gives us practical stuff to resolve. And I should say, it’s not that obvious to deal with all these formulas, you’d better be good in maths \:\)
10.00-12.00 The time has come for Grammatical Formalisms course, we can forget for a while about statistics and step into a more symbolical NLP class. And honestly, I like grammars \:\) There is something fascinating about all these different ways to manipulate a sentence structure! Having a linguistic background, I’ve always dealt with grammars, but in a completely opposite manner. Here we should chase NPs (nominal phrases) and VPs (verbal phrases), adjoin parse-trees, and face feature equations. 
Today’s class is about TAG  - Tree Adjoining Grammar. For me it’s the simplest formalism, as we’ve already seen some of it, during the work at one of the projects this year. 
12.00-13.00 Lunch at the campus’ cafeteria. Well, it’s not the best place to eat, even around campus, but not having much time to go anywhere else, I brace myself to eat a panini (a kind of a hot sandwich) with chicken. It’s not that bad actually, but after weeks on paninis, I would rather go for something else! Coming to France, don’t forget, that it’s the country eating the most of sandwiches in the world. So, you’ll find them everywhere, from train stations to restaurants, hot, cold, with all kind of stuffing, and in all kinds of quality as well. Regretfully, sandwiches are not my thing. 
Lunch time is always a time of our group, when we can discuss everything, like movies watched on weekend or recent methods in syntactic parsing. Studying NLP in Nancy, I met people from all over the world, as this university program hosts Erasmus Mundus students, actually there are only 3 French students out of a group of 10. So, we’re never out of topics! After today’s almost 30-minutes discussion about our forthcoming trip to Luxembourg, we finally quit the cafeteria and head to the bus stop, because we have a work meeting with our software project tutor in a regional computer science lab. 
14.00-15.00 Project meeting. Studying NLP is about practicing in many ways, so, besides going to regular classes every day, we have specific projects in different disciplines. Today’s meeting concerns Applications for NLP class, and our small sub-group of 4 girls is working on the creation of the linguistics resources for a natural language generator. We show the code for our small grammar created within the TAG-formalism and discuss its flaws with our tutor.
15.00-17.00 Still in the lab, we have some time to discuss another project among us, in Corpus Linguistics this time. We are to elaborate a solution to dialogue act prediction, given a set of dialogues with a virtual agent. A classifier based on conditional random fields in proposed by one of the members. Seems to be interesting, we’re going to stick to it, I think. 
Finally going home. It was a long day. Bonne soirée à tous! (good evening to everybody) like they say in France.

Natalia Korchagina

About the author:
Natalia Korchagina received her BA degree in French and German linguistics from the Volgograd State Pedagogical University (Russia). Now she is a 2nd year Master’s student in the Department of Mathematics and Informatics at Université de Lorraine (Nancy, France).  Natalia is graduating from the University in June 2013 and is already open for a new challenge in the NLP field.
LinkedIn - NLPPeople

No comments:

Post a Comment