Skip to content

Latest commit

 

History

History
226 lines (136 loc) · 10.3 KB

README.md

File metadata and controls

226 lines (136 loc) · 10.3 KB
Raw

Table of Contents

Kairyou

Quickly preprocesses Japanese text using NLP/NER from SpaCy for Japanese translation or other NLP tasks.

Quick Start

To get started with Kairyou, install the package via pip:

pip install kairyou

You must also install the jp_core_news_lg model from spaCy. This can be done by running the following command:

for windows:

python -m spacy download ja_core_news_lg

for linux:

python3 -m spacy download ja_core_news_lg

Then, you can preprocess Japanese text by importing Kairyou and/or KatakanaUtil/Indexer:

from kairyou import Kairyou, KatakanaUtil, Indexer

Follow the usage examples provided in the Usage section for detailed instructions on preprocessing text and handling katakana.

Installation

Python 3.10+

Kairyou can be installed using pip:

pip install kairyou

This will install Kairyou along with its dependencies, including spaCy and a few other packages.

These are the dependencies/requirements that will be installed:

setuptools>=61.0

wheel

setuptools_scm>=6.0

tomli

spacy==3.7.5

regex>=2023.3.23

Usage

Kairyou

Kairyou is the global preprocessor client. Here's an example of how to use it:

from kairyou import Kairyou

text = "Your Japanese text here."
replacement_json = "path/to/your/replacement_rules.json"  ## or a dict of rules
preprocessed_text, preprocessing_log, error_log = Kairyou.preprocess(text, replacement_json)

print(preprocessed_text)

Currently, Kairyou supports two json types, "Kudasai" and "Fukuin". "Kudasai" is the native type and originated from that program, Fukuin is what the original onegai program used, as well as what the kroatoan's Fukuin program uses. No major differences in replacement are present between the two.

Kairyou performs some post-processing on the text to correct any issues that may have arisen during the preprocessing.

Blank Kudasai Json

Example Kudasai Json

Blank Fukuin Json

Example Fukuin Json

KatakanaUtil

KatakanaUtil provides utility functions for handling katakana characters in Japanese text. Example usage:

from kairyou import KatakanaUtil

katakana_word = "カタカナ"
if KatakanaUtil.is_katakana_only(katakana_word):
    print(f"{katakana_word} is composed only of Katakana characters.")

The following functions are available in KatakanaUtil:

is_katakana_only: Returns True if the input string is composed only of katakana characters.

is_actual_word: Returns True if the input string is a actual Japanese Katakana word (not just something made up or a name). List of words can be found here.

is_punctuation: Returns True if the input string is punctuation (Both Japanese and English punctuation are supported). List of punctuation can be found here.

is_repeating_sequence: Returns True if the input string is just a repeating sequence of characters. (e.g. "ジロジロ")

is_more_punctuation_than_japanese: Returns True if the input string has more punctuation than Japanese characters.

is_partially_english: Returns True if the input string has any English characters in it.

Indexer

Indexer is for "indexing" Japanese text. What this means is that, given input_text, a knowledge_base, and a replacements_json. It will return a list of new "names", and the occurrence which was flagged.

What is considered a name is a bit complicated. But:

  1. Must have the "person" label when using spaCy's NER.
  2. Cannot have more punctuation than Japanese characters.
  3. Cannot be a repeating sequence of characters.
  4. Cannot be an actual Japanese Katakana word.
  5. Cannot have any english characters in it. (Names that are pre-replaced would have already been in the other texts, plus the JP NER model seems to have trouble with ENG in general)

So, it'll return names that don't in the other texts.

This can be done via index()

from kairyou import Indexer

input_text = "Your Japanese text here." ## or a path to a text file
knowledge_base = ["more Japanese text here.", "even_more_japanese_text_here"] ## or a path to a text file or directory full of text files
replacements_json = "path/to/your/replacement_rules.json"  ## or a dict of rules

## You can optionally send a list of strings to ignore, but this is not required.

NamesAndOccurrences, indexing_log = Indexer.index(input_text, knowledge_base, replacements_json) ## also takes in a list of strings to ignore, defaults to []

NamesAndOccurrences is a list of named tuples, with the following fields:

  1. name: The name that was found.
  2. occurrence: The occurrence of the name in the input_text.

Indexing_log is a string of the log of the indexing process. What was indexed, it's occurrence, what was ignored, and time elapsed.

Index works with both Fukuin and Kudasai jsons.

License

This project, Kairyou, is licensed under the GNU Lesser General Public License v2.1 (LGPLv2.1) - see the LICENSE file for complete details.

The LGPL is a permissive copyleft license that enables this software to be freely used, modified, and distributed. It is particularly designed for libraries, allowing them to be included in both open source and proprietary software. When using or modifying Kairyou, you can choose to release your work under the LGPLv2.1 to contribute back to the community or incorporate it into proprietary software as per the license's permissions.

Under the LGPLv2.1, any modifications made to Kairyou's libraries must be shared under the same license, ensuring the open source nature of the library itself is maintained. However, it allows the wider work that includes the library to remain under different licensing terms, provided the terms of the LGPLv2.1 are met for the library.

I encourage the use of Kairyou within all projects. This approach ensures a balance between open collaboration and the flexibility required for proprietary development.

For a thorough explanation of the conditions and how they may apply to your use of the project, please review the full license text. This comprehensive documentation will offer guidance on your rights and obligations when utilizing LGPLv2.1 licensed software.

Contact

If you have any questions or suggestions, feel free to reach out to me at Bikatr7@proton.me

Also feel free to check out the GitHub repository for this project.

Or the issue tracker here.

Contribution

Contributions are welcome! I don't have a specific format for contributions, but please feel free to submit a pull request or open an issue if you have any suggestions or improvements.

Notes

Kairyou was originally developed as a part of Kudasai, a Japanese preprocessor later turned Machine Translator. It was later split off into its own package to be used independently of Kudasai for multiple reasons.

Kairyou gets its name from the Japanese word "Reform" (改良) which is pronounced "Kairyou". Which was chosen for two reasons, the first being that it was chosen during a large Kudasai rework, and the second being that it is a Japanese preprocessor, and the name seemed fitting.

This package is also my first serious attempt at creating a Python package, so I'm sure there are some things that could be improved. Feedback is welcomed.

Inspirations

Kudasai and by extension Kairyou was originally derived from Void's Script later Onegai

Kairyou also took some inspiration from Fukuin and it's approach with Katakana.

Thanks to all of the above for the inspiration and the work they put into their projects.