yasuokaの日記: astremo/JAINUによる日本語→アイヌ語翻訳 1
1月29日の日記の読者から、astremo/JAINUという日本語→アイヌ語翻訳モデルを御教示いただいた。ちょっとGoogle Colaboratoryで試してみよう。
!pip install transformers sentencepiece
from transformers import AutoTokenizer,AutoModelForSeq2SeqLM,TranslationPipeline
tkz=AutoTokenizer.from_pretrained("astremo/JAINU")
mdl=AutoModelForSeq2SeqLM.from_pretrained("astremo/JAINU")
pipeline=TranslationPipeline(tokenizer=tkz,model=mdl)
print(pipeline("こんにちは"))
print(pipeline("さようなら"))
「こんにちは」と「さようなら」をアイヌ語に翻訳させてみたところ、私(安岡孝一)の手元では以下の結果になった。
[{'translation_text': 'ウオシ...オホシキノオ...オ...オ...オ...オ...オ...オ...オヤオ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オヤオホシキヤオシ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オヤオシ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オシ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オシ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オ...オシ...オ'}]
[{'translation_text': 'オラウンホシキチキ'}]
「さようなら」の方は「oraun hosi ki ciki」と翻訳されたようだが、私のアイヌ語の能力が低いのか、さっぱり意味がわからない。「こんにちは」の方は、かなり病的な事態になっており、翻訳モデルがマトモに動いていない気がする。現時点では、Language-Media-Labの各翻訳モデル(以前はgojitejiにあった)の方が精度がいいようだが、さて、今後どうなっていくだろう。
アイヌ語の保全に (スコア:1)