update April 2020

[Ted Pedersen]

Well it's been an unusual month or two to say the least, and I
realized I have not been as faithful in updating the NLP @ Duluth
discussion group as I had intended. I thought as we approach the end
of the spring semester it might be a good time to share a few updates.

I've been teaching a new Advanced NLP class semester, which has been
an interesting experience. The goal has been to be a Deep Learning in
NLP class, so we started with traditional Neural Nets and then moved
on to RNNs, Attention, Transformers, and the ubiquitous BERT family.
Much of what we are covering was unknown even a few years ago, and I
wonder what will have changed by the time the class is offered again
next Spring. The outline of the course can be found here, and
suggestions are always appreciated:

https://sites.google.com/d.umn.edu/umd-cs-5642-adv-nlp/assigned-reading-and-other-activities

In other news, I participated in SemEval-2020 Task 12
aka OffensEval-2020 (on my own) and in SemEval-2020 Task 7 (Assessing
the Funniness of Edited News Headlines) with a group that grew to
include 3 students.

My results in OffensEval are not terribly good. I started with a very
simple logistic regression approach which ended up being my final
approach. However, I don't mind the low ranking as I've been spending
some time working through the results and analyzing the training data.
The results from the Funny Headlines task were quite a bit better, and
utilized a BERT-style approach (in particular RoBERTa). Papers about
both are being written even now and I'll share those once they are
finalized. More info about each task can be found here :

OffenseEval : https://competitions.codalab.org/competitions/22917
Funny Headlines : https://competitions.codalab.org/competitions/20970

The Islamophobia project continues in a data collection and annotation
mode. I am hoping to have a paper describing some of that put together
this summer, although things are so unusual these days I am allowing
myself some flexibility with that goal.

As for conference "travel" I am planning to attend the virtual version
of ICLR (International Conference on Learning Representations) which
takes place at the end of April. There is an African NLP workshop I'm
interested in, and of late there have been lots of interesting Deep
Learning / NLP papers that appear there. This conference was going to
take place in Ethiopia but of course is virtual now.

https://iclr.cc/

I'll also attend the virtual ACL meeting which takes place in July
(formerly planned for Seattle). This is a very large international
gathering (Association for Computational Linguistics) and so I'm very
intrigued to see how this will come off in a virtual setting.

https://acl2020.org/

I don't have anything at either of these events so am just going to be
listening, learning, and hopefully meeting a few new people and also
connecting with some familiar faces as well. While virtual conferences
are not what anyone planned, it certainly makes it easier to
participate in these events. I was certainly not going to be able to
fly to Ethiopia for ICLR for example.

Finally, next week Saptarshi will be defending his project work, so
look for a separate announcement about that. This will of course be a
virtual event so you are all invited. :)

In any case, that's a bit of news from Duluth. I hope you are well
regardless of where this note finds you, and if you have your own news
or updates I'm always interested in those, and feel free to share
things of general interest on our discussion group too.

Stay well,
Ted
---
Ted Pedersen
http://www.d.umn.edu/~tpederse