John Bauer

Week 53 in sutime

1

Reported by Vijay Raj: Test sentences executed on Dec 30th, on corenlp.run 1. I was working last week. DATE NER: 2020-W52 2. I will be working next week. DATE NER:...

Use of the dataset for training models in relation to the license

4

Do you have any expectations for how the GPL v3 license should apply to weights in a model trained using the data as part of a training set, with no...

EOF lookahead?

3

Are there any plans to do lookahead that can see EOF? There's a bug in one of Stanford NLP group's tokenization tools which would be easily fixed by such an...

Stanford Morphology class uses POS tags

1

I'm not sure how you used the lemma annotator for CoreNLP to test the lemmatizer, but the Morphology class definitely does use POS tags if available: https://github.com/stanfordnlp/CoreNLP/blob/main/src/edu/stanford/nlp/process/Morphology.java For example, the...

Currently the depparse training uses gold lemmas. This means that at runtime, the lemmas the parser sees for user text or ETE scoring are slightly different. 4% in the case...

Unknown words can still result in punct tag at end of sentence

2

For example, if a sentence ends with `thicc` and no sentence final punctutation, `thicc` is labeled `PUNCT`

When using an English tokenizer with MWT (such as EWT, which treats possessive clitics as MWT), the NER model puts the tags on the entire token, not just the word,...

NER models start from a pretrained embedding, same as depparse, pos, etc however, they finetune the embedding the entire finetuned embedding is then saved in the model. only the delta...

draw_polygon does not fill in a square as I would expect

2

If I call `draw_polygon` with a square, it leaves the top row unfilled. Here is a small example: ``` >>> offset_simplex = numpy.array([[0, 4], [4, 4], [4, 0], [0, 0]])...