rhoknp: Yet another Python binding for Juman++/KNP/KWJA

Documentation: https://rhoknp.readthedocs.io/en/latest/

Source Code: https://github.com/ku-nlp/rhoknp

rhoknp is a Python binding for Juman++, KNP, and KWJA.¹

import rhoknp

# Perform morphological analysis by Juman++
jumanpp = rhoknp.Jumanpp()
sentence = jumanpp.apply_to_sentence(
    "電気抵抗率は電気の通しにくさを表す物性値である。"
)

# Access to the result
for morpheme in sentence.morphemes:  # a.k.a. keitai-so
    ...

# Save the result
with open("result.jumanpp", "wt") as f:
    f.write(sentence.to_jumanpp())

# Load the result
with open("result.jumanpp", "rt") as f:
    sentence = rhoknp.Sentence.from_jumanpp(f.read())

Requirements

Python 3.8+
(Optional) Juman++ v2.0.0-rc3+
(Optional) KNP 5.0+
(Optional) KWJA 1.0.0+

Installation

pip install rhoknp

Quick tour

Let's begin by using Juman++ with rhoknp. Here, we present a simple example demonstrating how Juman++ can be used to analyze a sentence.

# Perform morphological analysis by Juman++
jumanpp = rhoknp.Jumanpp()
sentence = jumanpp.apply_to_sentence("電気抵抗率は電気の通しにくさを表す物性値である。")

You can easily access the individual morphemes that make up the sentence.

for morpheme in sentence.morphemes:  # a.k.a. keitai-so
    ...

Sentence objects can be saved in the JUMAN format.

# Save the sentence in the JUMAN format
with open("sentence.jumanpp", "wt") as f:
    f.write(sentence.to_jumanpp())

# Load the sentence
with open("sentence.jumanpp", "rt") as f:
    sentence = rhoknp.Sentence.from_jumanpp(f.read())

Almost the same APIs are available for KNP.

# Perform language analysis by KNP
knp = rhoknp.KNP()
sentence = knp.apply_to_sentence("電気抵抗率は電気の通しにくさを表す物性値である。")

KNP performs language analysis at multiple levels.

for clause in sentence.clauses:  # a.k.a., setsu
    ...
for phrase in sentence.phrases:  # a.k.a. bunsetsu
    ...
for base_phrase in sentence.base_phrases:  # a.k.a. kihon-ku
    ...
for morpheme in sentence.morphemes:  # a.k.a. keitai-so
    ...

Sentence objects can be saved in the KNP format.

# Save the sentence in the KNP format
with open("sentence.knp", "wt") as f:
    f.write(sentence.to_knp())

# Load the sentence
with open("sentence.knp", "rt") as f:
    sentence = rhoknp.Sentence.from_knp(f.read())

Furthermore, rhoknp provides convenient APIs for document-level language analysis.

document = rhoknp.Document.from_raw_text(
    "電気抵抗率は電気の通しにくさを表す物性値である。単に抵抗率とも呼ばれる。"
)
# If you know sentence boundaries, you can use `Document.from_sentences` instead.
document = rhoknp.Document.from_sentences(
    [
        "電気抵抗率は電気の通しにくさを表す物性値である。",
        "単に抵抗率とも呼ばれる。",
    ]
)

Document objects can be handled in a similar manner as Sentence objects.

# Perform morphological analysis by Juman++
document = jumanpp.apply_to_document(document)

# Access language units in the document
for sentence in document.sentences:
    ...
for morpheme in document.morphemes:
    ...

# Save language analysis by Juman++
with open("document.jumanpp", "wt") as f:
    f.write(document.to_jumanpp())

# Load language analysis by Juman++
with open("document.jumanpp", "rt") as f:
    document = rhoknp.Document.from_jumanpp(f.read())

For more information, please refer to the examples and documentation.

Main differences from pyknp

pyknp serves as the official Python binding for Juman++ and KNP. In the development of rhoknp, we redesigned the API, considering the current use cases of pyknp. The key differences between the two are as follows:

Support for document-level language analysis: rhoknp allows you to load and instantiate the results of document-level language analysis, including cohesion analysis and discourse relation analysis.
Strict type-awareness: rhoknp has been thoroughly annotated with type annotations, ensuring strict type checking and improved code clarity.
Comprehensive test suite: rhoknp is extensively tested with a comprehensive test suite. You can view the code coverage report on Codecov.

License

MIT

Contributing

We warmly welcome contributions to rhoknp. You can get started by reading the contribution guide.

Reference

The logo was generated by OpenAI DALL·E 2. ↩

ku-nlp/rhoknp