You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Dataset Card for [MultiCCAligned-TW-Corpus]

Dataset Summary

MultiCCAligned-TW-Corpus 是一個機器翻譯基準的多語言資料集,源自 OPUS 收集的使用者貢獻的翻譯,並由 OPUS。該資料集包括按語言對排序的測試和開發資料。它包括數百種語言對的測試集,並且不斷更新。請檢查版本號標籤以引用您正在使用的版本。

Supported Tasks and Leaderboards

Languages

此資料集涵蓋數百種語言和語言對,並按 ISO-639-1 語言組織。目前版本涵蓋以下語言。繁體中文、英文、日文、韓文、印尼文、越南文、泰文

Dataset Structure

Data Instances

資料以 , 分隔檔案中內容,具有三個欄位:指示、輸入和輸出。請注意,我們並不暗示平移方向,並認為資料集是對稱的並用作兩個方向的測試集。

Data Splits

先整理出 Train 資料。

Dataset Creation

Curation Rationale

本資料集將持續更新,未來將公開發佈於 Github 當中。高語言覆蓋率是本計畫的主要目標,資料集的準備與標準化語言標籤和分發格式保持一致和系統化。

Source Data

Initial Data Collection and Normalization

MultiCCAligned 資料集是從根據 68 個 Commoncrawl 快照創建的(截至 2020 年 3 月)。根據標點符號將文件分割成句子,並執行重複資料刪除。語料庫的準備工作並沒有提出任何智慧財產權主張。原始發行版可從 http://www.statmt.org/cc-aligned/ 取得

Who are the source language producers?

這些轉錄本已由 EMNLP'20 作者群製作。

Personal and Sensitive Information

有關處理個人資訊和敏感資訊的信息,我們請諮詢資料的原始提供者。該資料集未經過任何方式處理以檢測或刪除潛在的敏感資訊或個人資訊。

Social Impact of Dataset

語言覆蓋率很高,因此它代表了機器翻譯開發的非常有價值的資源,特別是對於資源較少的語言和語言對。不斷成長的資料庫也代表著一種動態資源,其價值將進一步成長。

Other Known Limitations

這些句子通常很短,因此很容易翻譯。對於高資源語言,這會導致結果不如更具挑戰性的基準有用。對於資源較少的語言對來說,即使在非常具有挑戰性的設定中,範例的有限複雜性實際上也是衡量進度的一件好事。

Dataset Curators

此資料集由Heng-Shiou Sheu 製作。

Licensing Information

這些資料集沒使用 License.

Citation Information

@inproceedings{Heng666/MultiCCAligned-TW-Corpus,
  title={Taiwanese Phrases Multilingual Translation Dataset from MultiCCAligned Talks},
  author={Heng-Shiou Sheu},
  year={2024},
  url={https://huggingface.co/datasets/Heng666/MultiCCAligned-TW-Corpus},
}
Downloads last month
12
Edit dataset card

Collection including Heng666/MultiCCAligned-TW-Corpus