taibun 1.1.7

Taiwanese Hokkien Transliterator and Tokeniser

台語 | 國語

Taibun

Taiwanese Hokkien Transliterator and Tokeniser

It has methods that allow to customise transliteration and retrieve any necessary information about Taiwanese Hokkien pronunciation.
Includes word tokeniser for Taiwanese Hokkien.

Report Bug • PyPI

---

Table of Contents

1. Versions
2. Install
3. Usage
   • Converter
     • System
     • Dialect
     • Format
     • Delimiter
     • Sandhi
     • Punctuation
     • Convert non-CJK
   • Tokeniser
     • Keep original
   • Other Functions
4. Example
5. Data
6. Acknowledgements
7. Licence

Versions

Install

Taibun can be installed from pypi

$ pip install taibun

Usage

Converter

Converter class transliterates the Chinese characters to the chosen transliteration system with parameters specified by the developer. Works for both Traditional and Simplified characters.

# Constructor
c = Converter(system, dialect, format, delimiter, sandhi, punctuation, convert_non_cjk)

# Transliterate Chinese characters
c.get(input)

System

system String - system of transliteration.

• Tailo (default) - Tâi-uân Lô-má-jī Phing-im Hong-àn
• POJ - Pe̍h-ōe-jī
• Zhuyin - Taiwanese Phonetic Symbols
• TLPA - Taiwanese Language Phonetic Alphabet
• Pingyim - Bbánlám Uē Pìngyīm Hōng'àn
• Tongiong - Daī-ghî Tōng-iōng Pīng-im
• IPA - International Phonetic Alphabet

text	Tailo	POJ	Zhuyin	TLPA	Pingyim	Tongiong	IPA
台灣	Tâi-uân	Tâi-oân	ㄉㄞˊ ㄨㄢˊ	Tai5 uan5	Dáiwán	Tāi-uǎn	Tai²⁵ uan²⁵

Dialect

dialect String - preferred pronunciation.

• south (default) - Zhangzhou-leaning pronunciation
• north - Quanzhou-leaning pronunciation
• singapore - Quanzhou-leaning pronunciation with Singaporean characteristics

text	south	north	singapore
五月節我啉咖啡	Gōo-gue̍h-tseh guá lim ka-pi	Gōo-ge̍h-tsueh guá lim ka-pi	Gōo-ge̍h-tsueh uá lim ko-pi

Format

format String - format in which tones will be represented in the converted sentence.

• mark (default) - uses diacritics for each syllable. Not available for TLPA
• number - add a number which represents the tone at the end of the syllable
• strip - removes any tone marking

text	mark	number	strip
台灣	Tâi-uân	Tai5-uan5	Tai-uan

Delimiter

delimiter String - sets the delimiter character that will be placed in between syllables of a word.

Default value depends on the chosen system:

• '-' - for Tailo, POJ, Tongiong
• '' - for Pingyim
• ' ' - for Zhuyin, TLPA, IPA

text	'-'	''	' '
台灣	Tâi-uân	Tâiuân	Tâi uân

Sandhi

sandhi String - applies the sandhi rules of Taiwanese Hokkien.

Since it's difficult to encode all sandhi rules, Taibun provides multiple modes for sandhi conversion to allow for customised sandhi handling.

• none - doesn't perform any tone sandhi
• auto - closest approximation to full correct tone sandhi of Taiwanese, with proper sandhi of pronouns, suffixes, and words with 仔
• exc_last - changes tone for every syllable except for the last one
• incl_last - changes tone for every syllable including the last one

Default value depends on the chosen system:

• auto - for Tongiong
• none - for Tailo, POJ, Zhuyin, TLPA, Pingyim, IPA

text	none	auto	exc_last	incl_last
這是你的茶桌仔無	Tse sī lí ê tê-toh-á bô	Tse sì li ē tē-to-á bô	Tsē sì li ē tē-tó-a bô	Tsē sì li ē tē-tó-a bō

Sandhi rules also change depending on the dialect chosen.

text	no sandhi	south	north / singapore
台灣	Tâi-uân	Tāi-uân	Tài-uân

Punctuation

punctuation String

• format (default) - converts Chinese-style punctuation to Latin-style punctuation and capitalises words at the beginning of each sentence
• none - preserves Chinese-style punctuation and doesn't capitalise words at the beginning of new sentences

text	format	none
這是臺南，簡稱「南」（白話字：Tâi-lâm；注音符號：ㄊㄞˊ ㄋㄢˊ，國語：Táinán）。	Tse sī Tâi-lâm, kán-tshing "lâm" (Pe̍h-uē-jī: Tâi-lâm; tsù-im hû-hō: ㄊㄞˊ ㄋㄢˊ, kok-gí: Táinán).	tse sī Tâi-lâm，kán-tshing「lâm」（Pe̍h-uē-jī：Tâi-lâm；tsù-im hû-hō：ㄊㄞˊ ㄋㄢˊ，kok-gí：Táinán）。

Convert non-CJK

convert_non_cjk Boolean - defines whether or not to convert non-Chinese words. Can be used to convert Tailo to another romanisation system.

• True - convert non-Chinese character words
• False (default) - convert only Chinese character words

text	False	True
我食pháng	ㆣㄨㄚˋ ㄐㄧㄚㆷ˙ pháng	ㆣㄨㄚˋ ㄐㄧㄚㆷ˙ ㄆㄤˋ

Tokeniser

Tokeniser class performs NLTK wordpunct_tokenize-like tokenisation of a Taiwanese Hokkien sentence.

# Constructor
t = Tokeniser(keep_original)

# Tokenise Taiwanese Hokkien sentence
t.tokenise(input)

Keep original

keep_original Boolean - defines whether the original characters of the input are retained.

• True (default) - preserve original characters
• False - replace original characters with characters defined in the dataset

text	True	False
臺灣火鸡肉饭	['臺灣', '火鸡肉饭']	['台灣', '火雞肉飯']

Other Functions

Handy functions for NLP tasks in Taiwanese Hokkien.

to_traditional function converts input to Traditional Chinese characters that are used in the dataset. Also accounts for different variants of Traditional Chinese characters.

to_simplified function converts input to Simplified Chinese characters.

is_cjk function checks whether the input string consists entirely of Chinese characters.

to_traditional(input)

to_simplified(input)

is_cjk(input)

Example

# Converter
from taibun import Converter

## System
c = Converter() # Tailo system default
c.get('先生講，學生恬恬聽。')
>> Sian-sinn kóng, ha̍k-sing tiām-tiām thiann.

c = Converter(system='Zhuyin')
c.get('先生講，學生恬恬聽。')
>> ㄒㄧㄢ ㄒㆪ ㄍㆲˋ, ㄏㄚㆶ˙ ㄒㄧㄥ ㄉㄧㆰ˫ ㄉㄧㆰ˫ ㄊㄧㆩ.

## Dialect
c = Converter() # south dialect default
c.get("我欲用箸食魚")
>> Guá beh īng tī tsia̍h hî

c = Converter(dialect='north')
c.get("我欲用箸食魚")
>> Guá bueh īng tū tsia̍h hû

c = new Converter({ dialect: 'singapore' });
c.get("我欲用箸食魚");
>> Uá bueh ēng tū tsia̍h hû

## Format
c = Converter() # for Tailo, mark by default
c.get("生日快樂")
>> Senn-ji̍t khuài-lo̍k

c = Converter(format='number')
c.get("生日快樂")
>> Senn1-jit8 khuai3-lok8

c = Converter(format='strip')
c.get("生日快樂")
>> Senn-jit khuai-lok

## Delimiter
c = Converter(delimiter='')
c.get("先生講，學生恬恬聽。")
>> Siansinn kóng, ha̍ksing tiāmtiām thiann.

c = Converter(system='Pingyim', delimiter='-')
c.get("先生講，學生恬恬聽。")
>> Siān-snī gǒng, hág-sīng diâm-diâm tinā.

## Sandhi
c = Converter() # for Tailo, sandhi none by default
c.get("這是你的茶桌仔無")
>> Tse sī lí ê tê-toh-á bô

c = Converter(sandhi='auto')
c.get("這是你的茶桌仔無")
>> Tse sì li ē tē-to-á bô

c = Converter(sandhi='exc_last')
c.get("這是你的茶桌仔無")
>> Tsē sì li ē tē-tó-a bô

c = Converter(sandhi='incl_last')
c.get("這是你的茶桌仔無")
>> Tsē sì li ē tē-tó-a bō

## Punctuation
c = Converter() # format punctuation default
c.get("太空朋友，恁好！恁食飽未？")
>> Thài-khong pîng-iú, lín-hó! Lín tsia̍h-pá buē?

c = Converter(punctuation='none')
c.get("太空朋友，恁好！恁食飽未？")
>> thài-khong pîng-iú，lín-hó！lín tsia̍h-pá buē？

## Convert non-CJK
c = Converter(system='Zhuyin') # False convert_non_cjk default
c.get("我食pháng")
>> ㆣㄨㄚˋ ㄐㄧㄚㆷ˙ pháng

c = Converter(system='Zhuyin', convert_non_cjk=True)
c.get("我食pháng")
>> ㆣㄨㄚˋ ㄐㄧㄚㆷ˙ ㄆㄤˋ


# Tokeniser
from taibun import Tokeniser

t = Tokeniser()
t.tokenise("太空朋友，恁好！恁食飽未？")
>> ['太空', '朋友', '，', '恁好', '！', '恁', '食飽', '未', '？']

## Keep Original
t = Tokeniser() # True keep_original default
t.tokenise("爲啥物臺灣遮爾好？")
>> ['爲啥物', '臺灣', '遮爾', '好', '？']

t.tokenise("为啥物台湾遮尔好？")
>> ['为啥物', '台湾', '遮尔', '好', '？']

t = Tokeniser(False)
t.tokenise("爲啥物臺灣遮爾好？")
>> ['為啥物', '台灣', '遮爾', '好', '？']

t.tokenise("为啥物台湾遮尔好？")
>> ['為啥物', '台灣', '遮爾', '好', '？']


# Other Functions
from taibun import to_traditional, to_simplified, is_cjk

## to_traditional
to_traditional("我听无台语")
>> 我聽無台語

to_traditional("我爱这个个人台面")
>> 我愛這个個人檯面

to_traditional("爲啥物")
>> 為啥物

## to_simplified
to_simplified("我聽無台語")
>> 我听无台语

## is_cjk
is_cjk('我食麭')
>> True

is_cjk('我食pháng')
>> False

Data

• Taiwanese-Chinese Online Dictionary (via ChhoeTaigi)
• iTaigi Chinese-Taiwanese Comparison Dictionary (via ChhoeTaigi)

Acknowledgements

• Samuel Jen (Github · LinkedIn) - Taiwanese and Mandarin translation

Licence

Because Taibun is MIT-licensed, any developer can essentially do whatever they want with it as long as they include the original copyright and licence notice in any copies of the source code. Note, that the data used by the package is licensed under a different copyright.

The data is licensed under CC BY-SA 4.0