search

StrombergNLP / bornholmsk

NLP tools / data for Bornholmsk, NODALIDA 2019
https://stromberg.ai/publication/bornholmsknaturallanguageprocessingresourcesandtools/
Creative Commons Attribution 4.0 International
2 stars 0 forks source link

handle source-OOV words in bilingual #1

Open leondz opened 5 years ago

leondz commented 5 years ago

If we have a bilingual word pair and only have a vector for the target: Just map the word straight over, using the co-ordinates in the target space

leondz commented 5 years ago

consider adding a small amount of noise (mean 0; variance k=0.01 - ideally k should scale up with dim count)