BPEmb: ng

Ndonga (ng) subword embeddings

Vocab size	vocab	model	25 dim	50 dim	100 dim	200 dim	300 dim
1000	vocab	model	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix
3000	vocab	model	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix	txt \| bin bokeh \| umap \| matrix

View Ndonga subword embeddings in TensorFlow Projector

2-D UMAP plots

Embedding matrix plots

Training corpus sample, encoded with different BPE vocabulary sizes

Vocab size	ngwiki sample
original	omuntu kehe oku na uuthemba wemanguluko iyomadhiladhilo neiuvo osho wo elongelokalunga; uuthemba mbuka owa kwatelela mo emanguluko iyokulundulula elon af:duisburg ar:دويسبورغ bg:дуйсбург br:duisburg ca:duisburg cs:duisburg cy:duisburg da:duisburg de:duisburg el:ντούισμπουργκ en:duisburg eo:duisburg e # uuthemba mbuka inau dhimbulukiwa uuna omuntu ta pewa egeelo kaali na sha neyono iyopapolitika nenge keyon'o iyaa na sha nelalakano nomakankameno gii
1000	▁omuntu ▁kehe ▁oku ▁na ▁uuthemba ▁wemanguluko ▁iyo madhiladhilo ▁neiuvo ▁osho ▁wo ▁elongelokalunga ; ▁uuthemba ▁mbuka ▁owa ▁kwatelela ▁mo ▁emanguluko ▁iyo kul un d ulu la ▁e lo n ▁a f : duisburg ▁ar : دو ي سبور غ ▁bg : дуйсбург ▁b r : duisburg ▁ca : duisburg ▁cs : duisburg ▁cy : duisburg ▁da : duisburg ▁de : duisburg ▁el : ντ ού ισμ πο υρ γκ ▁en : duisburg ▁eo : duisburg ▁e ▁# ▁uuthemba ▁mbuka ▁ina u ▁dhi mbulu ki wa ▁uuna ▁omuntu ▁ta ▁pewa ▁egeelo ▁kaa li ▁na ▁sha ▁ne yono ▁iyo pa politika ▁nenge ▁ke yo n ' o ▁iyaa ▁na ▁sha ▁ne lalakano ▁no ma kanka men o ▁g ii
3000	▁omuntu ▁kehe ▁oku ▁na ▁uuthemba ▁wemanguluko ▁iyomadhiladhilo ▁neiuvo ▁osho ▁wo ▁elongelokalunga ; ▁uuthemba ▁mbuka ▁owa ▁kwatelela ▁mo ▁emanguluko ▁iyokulundulula ▁e lon ▁af : duisburg ▁ar : دويسبورغ ▁bg : дуйсбург ▁br : duisburg ▁ca : duisburg ▁cs : duisburg ▁cy : duisburg ▁da : duisburg ▁de : duisburg ▁el : ντούισμπουργκ ▁en : duisburg ▁eo : duisburg ▁e ▁# ▁uuthemba ▁mbuka ▁inau ▁dhimbulukiwa ▁uuna ▁omuntu ▁ta ▁pewa ▁egeelo ▁kaali ▁na ▁sha ▁neyono ▁iyopapolitika ▁nenge ▁keyon ' o ▁iyaa ▁na ▁sha ▁nelalakano ▁nomakankameno ▁g ii