Aug 8, 2017 · 1 min read
This Hackaton looks like it was a great experience, the new features you are suggesting ( “Play some Pop Rock from the 90s)”are highly appreciable from the user point of view. And this is an interesting challenge from the NLP researcher point of view. I have experienced problems while trying to perform Named Entity Recognition on the Speech to text output, since the output is rarely capitalized - one of the most important features for good NER taggers. I’ve heard some taggers are inspired from social media data where capitalization is rarely respected as well. Did you find any solution for this yet?
