本項目為 canto-filter 之後續。canto-filter 得 4 個分類標籤且判斷邏輯更加快速簡單,適合在線快速篩選判別文本或者其他要求低延遲、速度快嘅應用場合。本項目採用更精細嘅判斷邏輯,有 6 個分類標籤,準確度更高,但速度亦會相對 canto-filter 更慢。
This is an extension of the canto-filter project. canto-filter has only 4 output labels. It has a simipler classification logic and is faster, more suitable for use cases which require low-latency or high classification speed. This package has 6 output and uses a more sophisticated classification logic for more fine-grained classification. It has higher classification accuracy but slower performance.
分類器採用嘅分類標籤及基準,參考咗對使用者嘅語言意識形態嘅研究。討論分類準則時,請引用:
The definitions and boundaries of the labels depend on the user's language ideology. When discussing the criteria adopted by this tool, please cite:
Chaak-ming Lau, Mingfei Lau, and Ann Wai Huen To. 2024. The Extraction and Fine-grained Classification of Written Cantonese Materials through Linguistic Feature Detection. In Proceedings of the 2nd Workshop on Resources and Technologies for Indigenous, Endangered and Lesser-resourced Languages in Eurasia (EURALI) @ LREC-COLING 2024, pages 24–29, Torino, Italia. ELRA and ICCL.
分類方法係利用粵語同書面中文嘅特徵字詞,用 Regex 方式加以識別。分類器主要有兩個主要參數,--split
同埋--quotes
,兩個默認都係False
。
The filter is based on Regex rules and detects lexical features specific to Cantonese or Written-Chiense.
呢個參數默認關閉,如果打開,分類器會用句號、問號、感歎號等標點符號將輸入文本切成單句,對每個單句分類判斷,然後再按照下面判別標準整合嚟得到最終分類。所以呢個參數喺輸入都係單句嘅情況下唔會有區別,只會降低運行速度。喺官粵混雜比較多而且比較長嘅文本輸入下會有更多唔同。
目前因為整合分句判斷嘅邏輯比較嚴,所以如果打開,會相比於關閉更加容易將其他類別判斷為mixed
。所以對於篩選純粵文嘅用途嚟講,打開呢個參數會提高 precision 但降低 recall。
呢個參數默認關閉,分類器淨係會將輸入分為 4 類。如果打開,就會再增加兩類總共有 6 個標籤。打開後分類器會將引號內嘅文本抽出嚟,將佢哋同引號外文本分開判斷。下面一段就係介紹呢四個同六個標籤。
分類器會將輸入文本分成四類(粗疏)或六類(精細),分類如下:
The classifiers output four (coarse) or six (fine-grained) categories. The labels are:
Cantonese
: 純粵文,僅含有粵語特徵字詞,例如“你喺邊度” | Pure Cantonese text, contains Cantonese-featured words. E.g. 你喺邊度SWC
: 書面中文,係一個僅含有書面語特徵字詞,例如“你在哪裏” | Pure Standard Written Chinese (SWC) text, contains Mandarin-feature words. E.g. 你在哪裏Mixed
:書粵混雜文,同時含有書面語同粵語特徵嘅字詞,例如“是咁的” | Mixed Cantonese-Mandarin text, contains both Cantonese and Mandarin-featured words. E.g. 是咁的Neutral
:無特徵中文,唔含有官話同粵語特徵,既可以當成粵文亦可以當成官話文,例如“去學校讀書” | No feature Chinese text, contains neither Cantonese nor Mandarin feature words. Such sentences can be used for both Cantonese and Mandarin text corpus. E.g. 去學校讀書MixedQuotesInSWC
: 書面中文,引文入面係Mixed
|Mixed
contents quoted within SWC textCantoneseQuotesInSWC
: 書面中文,引文入面係純粵文cantonese
|Cantonese
contents quoted within SWC text
Python >= 3.11
pip install cantonesedetect
可以通過 Python 函數嚟引用,亦可以直接 CLI 調用。
You can call the Python API or this library, or run it directly in CLI.
用下面嘅方法創建一個 Detector
,然後直接調用 judge()
就可以得到分類結果:
Initialize a Detector
and call the judge()
function on inputs, and you will get the classification outputs.
from cantonesedetect import CantoneseDetector
# 默認情況下 use_quotes=False, split_seg=False, get_analysis=False
detector = CantoneseDetector()
detector.judge('你喺邊度') # cantonese
detector.judge('你在哪裏') # swc
detector.judge('是咁的') # mixed
detector.judge('去學校讀書') # neutral
detector.judge('他説:“係噉嘅。”') # cantonese_quotes_in_swc
detector.judge('那就「是咁的」') # mixed_quotes_in_swc
如果想要用引號抽取判別、分句判別同埋獲得分析結果,可以:
If you want to judge inputs based on matrix-quote-splitting, or spliting into segments, you can:
from cantonesedetect import Detector
detector = Detector(use_quotes=True, split_seg=True, get_analysis=True)
judgement, document_features = detector.judge("他説:「我哋今晚食飯。你想去邊度食?」")
# 打印分析結果
# Print analysis results
print(document_features.get_analysis())
# `document_features` 入面有每個分句嘅 `document_segments_features` 同 `document_segments_judgements`
# `document_features` object contains `document_segments_features` which is a list of segment features
print(document_features.document_segments_features[0].canto_feature)
print(document_features.document_segments_features[0].canto_exclude)
print(document_features.document_segments_features[0].swc_feature)
print(document_features.document_segments_features[0].swc_exclude)
# Also contains `document_segments_judgements` which is a list of judgements of the segments
print([j.value for j in document_features.document_segments_judgements])
如果直接喺 CLI 調用嘅話,只需要指明--input
就得。 --quotes
、--split
、--print_analysis
三個參數都默認關閉,如果標明就會打開:
If you run directly in CLI, simply specify the --input
. The optional arguments --quotes
、--split
、--print_analysis
are all False
by default, and you can turn them on by specifying them.
cantonesedetect --input input.txt
# 開啓引號抽取判別、分句判別並且打印分析結果
# Enable matrix-quotes-splitting, segment-splitting and printing the analysis.
cantonesedetect --input input.txt --quotes --split --print_analysis