#introductions Hey everyone. I am a PhD Student in Computational Linguistics. My main research interest is computational linguistics for under resourced languages. I am particularly interested in computational linguistics for Bantu languages because of the extremely complex morphology that they bring to the table.
I am also doing research in sentiment analysis on stance detection and irony/sarcasm detection. Most of my research is on machine learning approaches.
You know, for an animal who sleeps 80% of the time, my cat is not very accepting of me taking a nap.
#introductions Hi, my name is Harm, a strange-sounding name for English-speaking people, but it will have to do. I'm a 50 y.o., upper middle class, cis/he/hetero, happily married, white father of one, doing a PhD on #Hegel at Utrecht University. I'm self-employed as a private philosophical tutor (some would say 'counselor', but I don't give counsel), teacher, and writer. I'm far left politically. I sing the praises of good wine, high art, and beautiful women. I sin boldly. Nice to meet you.
Looks like it's making more sense to buy a better gpu if I want to figure out how to train morphological analysis models.
#Google's cloud provisioning stuff seems much more difficult to start out with than Amazon's. In addition, it is requiring me to link my account to the university's organization which isn't ideal since it says that the domain administrators have access to all the resources for all the projects I create.
I also keep hitting broken pages and the ui is wayy over the top for what I need to do (just create a vm with a gpu or tpu).
Gpu time is too expensive on aws and Google"s cloud is too clunky
Reading Evaluation Learning Algorithms: a classification perspective and I came across this:
"The use of cross-validation causes accrued uncertainty in the ensuing t-test because the learned classifieds are not independent of each other." (2011, p 14)
I wonder if the uncertainty in the t-test rises as the number of folds does. Seems leave-one-out would be the worst offender. However, I'm not sure if the uncertainty is actually something that can have values here or if independence violated.
my landlord rocks Show more
Man... I really lucked out that I don't have Hallmark Rentals for property management on the place I'm moving to.
They took weeks to answer my emails, didn't answer my calls and charged me dumb fees like a Re-keying fee.
Now the owner of the property is self managing and he's been great! He answers my dumb questions very quickly, he's been really cool about painting and flexible with move-in scheduling.
"The comments were surprisingly civil and thoughtful, but the article itself dismayed me. Croshaw clearly doesn't understand the importance of other perspectives--nonwhite, nonmale perspectives especially."
YouTube link Show more
This is hilarious http://www.youtube.com/watch?v=4Rz8BPRFeiA&t=3m22s
Finished enrolling in classes for the fall. I only needed one comp Sci course to finish out my minor and I found this cool music informatics course that I'm now really looking forward to
I’m looking for public domain comic books in a venue that allows me to batch download.
I usually use comic book plus, but they limit you to one download every two minutes.
I could bypass that, but I’d rather not.
I have a thing I’m working on that would be easier if I could pre-download all the comics I want to use, in a batch while I’m doing other stuff.
I suspect that meta information is very predictive of how many citations a paper is going to receive. If Joachim Nivre writes a paper, it will be highly cited.
I also suspect that there's institutional effects. A researcher that brings in a large grant for a particular project is going to bias the results. Their university will become more predictive of citation counts since there is likely a large amount of well-funded research being generated as a result of that grant.This is hypothetical tho
Impact paper from LREC is up in the proceedings now: http://lrec-conf.org/workshops/lrec2018/W6/pdf/2_W6.pdf .
Not the best paper ever but it's a decent first step into determining what makes a successful paper in computational linguistics.
Next steps on this project is to see whether meta information like University or Author is more predictive than just raw textual data.
The implications of this next stage of investigation are a lot more high stakes however.