PORORO: Platform Of neuRal mOdels for natuRal language prOcessing
pororo performs Natural Language Processing and Speech-related tasks.
It is easy to solve various subtasks in the natural language and speech processing field by simply passing the task name.
Installation
-
pororois based ontorch=1.6(cuda 10.1)andpython>=3.6 -
You can install a package through the command below:
pip install pororo
- Or you can install it locally:
git clone https://github.com/kakaobrain/pororo.git
cd pororo
pip install -e .
-
For library installation for specific tasks other than the common modules, please refer to INSTALL.md
-
For the utilization of Automatic Speech Recognition, wav2letter should be installed separately. For the installation, please run the asr-install.sh file
bash asr-install.sh
Usage
pororocan be used as follows:- First, in order to import
pororo, you must execute the following snippet
>>> from pororo import Pororo
- After the import, you can check the tasks currently supported by the
pororothrough the following commands
>>> from pororo import Pororo
>>> Pororo.available_tasks()
"Available tasks are ['mrc', 'rc', 'qa', 'question_answering', 'machine_reading_comprehension', 'reading_comprehension', 'sentiment', 'sentiment_analysis', 'nli', 'natural_language_inference', 'inference', 'fill', 'fill_in_blank', 'fib', 'para', 'pi', 'cse', 'contextual_subword_embedding', 'similarity', 'sts', 'semantic_textual_similarity', 'sentence_similarity', 'sentvec', 'sentence_embedding', 'sentence_vector', 'se', 'inflection', 'morphological_inflection', 'g2p', 'grapheme_to_phoneme', 'grapheme_to_phoneme_conversion', 'w2v', 'wordvec', 'word2vec', 'word_vector', 'word_embedding', 'tokenize', 'tokenise', 'tokenization', 'tokenisation', 'tok', 'segmentation', 'seg', 'mt', 'machine_translation', 'translation', 'pos', 'tag', 'pos_tagging', 'tagging', 'const', 'constituency', 'constituency_parsing', 'cp', 'pg', 'collocation', 'collocate', 'col', 'word_translation', 'wt', 'summarization', 'summarisation', 'text_summarization', 'text_summarisation', 'summary', 'gec', 'review', 'review_scoring', 'lemmatization', 'lemmatisation', 'lemma', 'ner', 'named_entity_recognition', 'entity_recognition', 'zero-topic', 'dp', 'dep_parse', 'caption', 'captioning', 'asr', 'speech_recognition', 'st', 'speech_translation', 'ocr', 'srl', 'semantic_role_labeling', 'p2g', 'aes', 'essay', 'qg', 'question_generation', 'age_suitability']"
- To check which models are supported by each task, you can go through the following process
>>> from pororo import Pororo
>>> Pororo.available_models("collocation")
'Available models for collocation are ([lang]: ko, [model]: kollocate), ([lang]: en, [model]: collocate.en), ([lang]: ja, [model]: collocate.ja), ([lang]: zh, [model]: collocate.zh)'
- If you want to perform a specific task, you can put the task name in the
taskargument and the language name in thelangargument
>>> from pororo import Pororo
>>> ner = Pororo(task="ner", lang="en")
- After object construction, it can be used in a way that passes the input value as follows:
>>> ner("Michael Jeffrey Jordan (born February 17, 1963) is an American businessman and former professional basketball player.")
[('Michael Jeffrey Jordan', 'PERSON'), ('(', 'O'), ('born', 'O'), ('February 17, 1963)', 'DATE'), ('is', 'O'), ('an', 'O'), ('American', 'NORP'), ('businessman', 'O'), ('and', 'O'), ('former', 'O'), ('professional', 'O'), ('basketball', 'O'), ('player', 'O'), ('.', 'O')]
- If task supports multiple languages, you can change the
langargument to take advantage of models trained in different languages.
>>> ner = Pororo(task="ner", lang="ko")
>>> ner("마이클 제프리 조던(영어: Michael Jeffrey Jordan, 1963년 2월 17일 ~ )은 미국의 은퇴한 농구 선수이다.")
[('마이클 제프리 조던', 'PERSON'), ('(', 'O'), ('영어', 'CIVILIZATION'), (':', 'O'), (' ', 'O'), ('Michael Jeffrey Jordan', 'PERSON'), (',', 'O'), (' ', 'O'), ('1963년 2월 17일 ~', 'DATE'), (' ', 'O'), (')은', 'O'), (' ', 'O'), ('미국', 'LOCATION'), ('의', 'O'), (' ', 'O'), ('은퇴한', 'O'), (' ', 'O'), ('농구 선수', 'CIVILIZATION'), ('이다.', 'O')]
>>> ner = Pororo(task="ner", lang="ja")
>>> ner("マイケル・ジェフリー・ジョーダンは、アメリカ合衆国の元バスケットボール選手")
[('マイケル・ジェフリー・ジョーダン', 'PERSON'), ('は', 'O'), ('、アメリカ合衆国', 'O'), ('の', 'O'), ('元', 'O'), ('バスケットボール', 'O'), ('選手', 'O')]
>>> ner = Pororo(task="ner", lang="zh")
>>> ner("麥可·傑佛瑞·喬丹是美國退役NBA職業籃球運動員，也是一名商人，現任夏洛特黃蜂董事長及主要股東")
[('麥可·傑佛瑞·喬丹', 'PERSON'), ('是', 'O'), ('美國', 'GPE'), ('退', 'O'), ('役', 'O'), ('nba', 'ORG'), ('職', 'O'), ('業', 'O'), ('籃', 'O'), ('球', 'O'), ('運', 'O'), ('動', 'O'), ('員', 'O'), ('，', 'O'), ('也', 'O'), ('是', 'O'), ('一', 'O'), ('名', 'O'), ('商', 'O'), ('人', 'O'), ('，', 'O'), ('現', 'O'), ('任', 'O'), ('夏洛特黃蜂', 'ORG'), ('董', 'O'), ('事', 'O'), ('長', 'O'), ('及', 'O'), ('主', 'O'), ('要', 'O'), ('股', 'O'), ('東', 'O')]
- If the task supports multiple models, you can change the
modelargument to use another model.
>>> from pororo import Pororo
>>> mt = Pororo(task="mt", lang="multi", model="transformer.large.multi.mtpg")
>>> fast_mt = Pororo(task="mt", lang="multi", model="transformer.large.multi.fast.mtpg")
Documentation
For more detailed information, see full documentation
If you have any questions or requests, please report the issue.
Citation
If you apply this library to any project and research, please cite our code:
@misc{pororo,
author = {Heo, Hoon and Ko, Hyunwoong and Kim, Soohwan and
Han, Gunsoo and Park, Jiwoo and Park, Kyubyong},
title = {PORORO: Platform Of neuRal mOdels for natuRal language prOcessing},
howpublished = {\url{https://github.com/kakaobrain/pororo}},
year = {2021},
}
Contributors
Hoon Heo, Hyunwoong Ko, Soohwan Kim, Gunsoo Han, Jiwoo Park and Kyubyong Park
License
PORORO project is licensed under the terms of the Apache License 2.0.
Copyright 2021 Kakao Brain Corp. https://www.kakaobrain.com All Rights Reserved.
7 Oct 30, 2022
4 Feb 27, 2022
26 Dec 05, 2022
36 Sep 13, 2022
136 Dec 20, 2022
0 Feb 06, 2022
40 Nov 07, 2022
6 Oct 18, 2022
98 Nov 24, 2022
1 Mar 04, 2022
3 Apr 19, 2022
30 Sep 07, 2022
179 Jan 07, 2023
21 Dec 14, 2022
2 Feb 17, 2022
22 Nov 16, 2022
544 Jan 09, 2023
30 Dec 13, 2022
587 Dec 20, 2022
57 Dec 07, 2022