Datasets:
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.
本資料集採人工審核制。請簡述你的使用目的與所屬單位; 僅供研究與模型訓練用途,請勿用於重製原始網路文本之散布。
Log in or Sign Up to review the conditions and access this dataset content.
YAML Metadata Warning:The task_categories "text2text-generation" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
zh-CN → zh-TW Localization Pairs
簡體中文 → 台灣繁體中文在地化平行語料:49,857 組句對,每組包含真實簡體中文 段落與對應的台灣繁中改寫(字形轉換 + 中國大陸用語 → 台灣慣用詞)。
- 輸出由內部的大型在地化模型生成(synthetic),非人工翻譯
- 每列附七道自動品質檢查的判定(
c_pass),整體通過率 87.8% - 用途:訓練/蒸餾輕量的簡轉繁在地化模型、簡繁轉換評測
欄位
| 欄位 | 型別 | 說明 |
|---|---|---|
id |
str | sha1(src) 前 16 碼,全集唯一 |
src |
str | 簡體中文原文(真實網路文本,非合成簡中) |
out |
str | 台灣繁中在地化輸出 |
simp_ratio |
float | 原文的簡體專屬字率(經驗字頻法判定) |
c_pass |
bool | 七道檢查全數通過 |
c_reasons |
str | 未通過時的原因(分號分隔) |
來源與生成
輸入:取自公開網路語料的真實簡體中文段落,以兩個判別器篩選: 簡體專屬字率 ≥ 5%(698 字經驗字頻集:簡中 ppm≥20 且繁中 ppm<0.5)、 壓縮率 ≤ 1.5 字/token。刻意不用「繁中反向合成的假簡中」—— 合成簡中的句法節奏與話題分布仍是原生繁中的,存在 domain gap。
輸出:由大型在地化模型以「在地化編輯」指令生成(溫度 0.2): 字形一律轉繁體、大陸慣用語改台灣慣用詞(信息→資訊、軟件→軟體、視頻→影片…)、 保持事實內容與段落結構、專有名詞僅做字形轉換。
七道品質檢查(c_pass 的定義)
| # | 檢查 | 門檻 | 未通過數 |
|---|---|---|---|
| 1 | 長度保真 | 0.90–1.10 | 879 |
| 2 | 陸用語殘留(deterministic 詞表,遮引號、含 cancel/exception) | 零命中 | 506 |
| 3 | 簡體字殘留 | < 0.5% | 863 |
| 4 | 前言(「好的,以下是…」) | 無 | 7 |
| 5 | 新增 markdown | 無 | 59 |
| 6 | 數字保真(全形正規化後多重集合相等) | 完全一致 | 4,287 |
| 7 | 異體字(爲/裏/着/温…) | 零 | 179 |
長度比分布 p10/p50/p90 = 0.965/0.996/1.028。
檢查 #6 偏嚴格(阿拉伯數字↔中文數字的合法改寫也會被刷),
需要最大量時可自行放寬;一般用途建議直接取 c_pass == true 的 43,754 列。
限制與注意
- 輸出是模型生成的合成資料,未經人工逐筆審核;建議在訓練語料中 控制其佔比,並混入原生繁中文本。
- 原文為真實網路文本,可能含其觀點、偏誤與時空脈絡(例:「北京時間」 在地化後仍是北京時間)。政治/社會類已盡量避開,但無法保證全無。
- 字形轉換之外的語義層歧義(質量/代碼等同形異義)由模型判斷, 正確率未逐類量測。
- 後續版本規劃:含 token 級 soft label 的蒸餾版、擴充至 20 萬組。
授權
ODC-BY 1.0(輸入承襲公開網路語料之授權脈絡)。使用本資料集請同時遵守 原始網路內容的相關權利規範。
- Downloads last month
- 12