koreannet | neural architecture for modeling the unique compositional

koreannet is a Perl library. koreannet has no bugs, it has no vulnerabilities and it has low support. You can download it from GitHub.

KoreanNet is a neural architecture for modeling the unique compositional orthography of the Korean language. It decomposes each character to extract underlying jamo letters (basic phonetic units) from which character represenations are composed. For example, the embedding for the character 갔 is a function of ㄱ, ㅏ, and ㅆ. The decomposition can be performed deterministically and efficiently by simple Unicode manipulation: we use a robust implementation provided here. This package plugs KoreanNet into the wonderful BiLSTM parser of Kiperwasser and Goldberg (2016). The jamo-based model achieves the high performance of character-based models while eschewing the need to store combinatorially many character types as lookup parameters. There is further improvement when jamos, characters, and words are used in conjunction. For details please refer to the paper.