Descritption

This repository contains a set of scripts to build a ready-to-use Juman++ model for Jumandic.

Prerequrements

• Unix environment (on Windows use WSL or MSYS2/MinGW64)
• Juman++ build environment
• Python 3.6+
• Ruby
• Perl
• Configured ssh authorization for github (we will clone several repositories via ssh)
• 32 GB of RAM

Recommended

• Original texts from Mainichi Shinbun (year 1995) for Kyoto Corpus (see the page for more information). Othewise, Juman++ model will be trained only on Leads corpus and will have poor quality.

How to Use

Run the configuration script: python3 configure.py. It will prompt for the location of Mainichi Shinbun texts.

After that run make nornn for training a model without RNN component. make rnn produces the model with RNN component. The models will be inside the bld/model folder.

Adding your words to the model

It is possible to add your words to the model. To do it:

1. Perform the configuration as described above: python3 configure.py
2. Fetch the repositories make repo.
3. Go into bld/repos/jumandic folder, it is a local clone of JumanDIC repository.
4. Create a new file with the .dic extension in the userdic folder of the bld/repos/jumandic folder.
5. Put your words into that file, in JUMAN dictionary format (refer to other files for example).
6. Execute make clean-dic if you have already built a Juman++ model.
7. Build your model as shown above.

If the built model does not contain your words, ensure that the binary dictionary was rebuilt after adding new words.