Support word and subword language models

In !274 (merged), we generate files for character LM during data extraction. We also need to generate the resources for word and subword LM.