Tokenization Subword ee Soomaaliga isku-dhafka ah

Dulucda daraasadda

Waxaan daraaseynay saamaynta BPE, Unigram, iyo tokenization morphology-aware ee hawlaha NLP Soomaaliga ee hoose, waxaana muujinaynaa in moodellada morphology-informed ay yareeyaan out-of-vocabulary 33% una hagaajiyaan classification F1 illaa 6 dhibcood.