Machine translation tools do not yet exist for the Yupík language, a polysynthetic language spoken by around 8,000 people who live primarily in Southwest Alaska. We compiled a parallel text corpus for Yupík and English and developed a morphological parser for Yupík based on grammar rules. We trained a seq2seq neural machine translation model with attention to translate Yupík input into English. We then compared the influence of different tokenization methods, namely rule-based, unsupervised (byte pair encoding), and unsupervised morphological (Morfessor) parsing, on BLEU score accuracy for Yupík to English translation. We find that using tokenized input increases the translation accuracy compared to that of unparsed input. Although overall Morfessor did best with a vocabulary size of 30k, our first experiments show that BPE performed best with a reduced vocabulary size.
View on arXiv