Existing work: Named entity recognition

[danieldk]

[danieldk]

sticker is a neural sequence labeler (bidi RNNs or dilated convolution networks) which could do named entity recognition. One of our students has done some experiments on the effect of word representations on named entity recognition. Has work shows that adding character LSTMs or convolutions would improve results over using subword embeddings (which we are currently using). Adding support for char RNNs and convolutions is on my todo list.

I have outlined a list of other items that we are working on in sticker in #6.