Yontenn commited on
Commit
c489ca7
·
verified ·
1 Parent(s): 86927a4

Card in the standard format: BDRC names, data source and preparation

Browse files
Files changed (1) hide show
  1. README.md +19 -7
README.md CHANGED
@@ -1,5 +1,13 @@
1
  ---
2
  license: other
 
 
 
 
 
 
 
 
3
  language:
4
  - bo
5
  task_categories:
@@ -41,17 +49,21 @@ configs:
41
  path: windowed_w8192_s4579_bioe_reviewed/test/*.parquet
42
  ---
43
 
44
- # Tibetan Quotation Detection Benchmark
 
 
45
 
46
  **254 classical Tibetan books** with **23,095 human-reviewed quotation spans**, split by book into train / validation / test.
47
 
48
  Character offsets are **inclusive on both ends**: the span is `text[start:end+1]`.
49
 
50
- ## Source
 
 
51
 
52
  Classical Buddhist commentaries digitized with support from the Tsadra Foundation and OpenPecha, annotated in two batches: an old batch (125 books, `Quotation` layer) and a new batch (195 books, `Citation` layer). Both layers mark the same thing — one text quoting a scripture, a named teacher, or a stock phrase — so they are merged into one label.
53
 
54
- ## Cleaning
55
 
56
  1. **Human review.** 398 books reviewed; 78 rejected whole for bad boundaries; 254 boundaries repaired in accepted books.
57
  2. **Span merging.** The new batch often splits one quotation across lines. Pieces separated only by 0–2 spaces, newlines, or a shad (and not already closed) were joined: 28,892 → 23,476 spans.
@@ -100,7 +112,7 @@ Books are **stratified by quote type** (`TEXT` / `PERSON` / `MISC`), so each spl
100
 
101
  ```python
102
  from datasets import load_dataset
103
- ds = load_dataset("karma689/tibetan-quotation-detection")
104
  ```
105
 
106
  ## Windowed configs (for mmBERT training)
@@ -117,7 +129,7 @@ All quote types collapse to one `QUOTE` label. Load the parquet files directly (
117
  ```python
118
  from datasets import load_dataset
119
  cfg = "windowed_w8192_s4579" # or windowed_w8192_s4579_bioe
120
- base = f"hf://datasets/karma689/tibetan-quotation-detection/{cfg}"
121
  ds = load_dataset("parquet", data_files={
122
  s: f"{base}/{s}/*.parquet" for s in ["train", "validation", "test"]
123
  })
@@ -148,7 +160,7 @@ The BIO config is also pinned at tag `w1.0`.
148
  title = {Tibetan Quotation Detection Benchmark Dataset},
149
  author = {karma689},
150
  year = {2026},
151
- url = {https://huggingface.co/datasets/karma689/tibetan-quotation-detection}
152
  }
153
  ```
154
 
@@ -189,7 +201,7 @@ Two new configs, both test-only:
189
 
190
  ```python
191
  from datasets import load_dataset
192
- test = load_dataset("karma689/tibetan-quotation-detection",
193
  data_files={"test": "windowed_w8192_s4579_bioe_reviewed/test/*.parquet"},
194
  revision="w2.1")["test"]
195
  ```
 
1
  ---
2
  license: other
3
+ pretty_name: Bo-Quotation-Dataset
4
+ tags:
5
+ - tibetan
6
+ - quotation
7
+ - bio
8
+ - bioe
9
+ - mmbert
10
+ - openpecha
11
  language:
12
  - bo
13
  task_categories:
 
49
  path: windowed_w8192_s4579_bioe_reviewed/test/*.parquet
50
  ---
51
 
52
+ # Bo-Quotation-Dataset
53
+
54
+ Benchmark and training data for finding **quotation spans** in classical Tibetan commentaries. It is the training data of [`BDRC/Bo-Quotation-Detection`](https://huggingface.co/BDRC/Bo-Quotation-Detection).
55
 
56
  **254 classical Tibetan books** with **23,095 human-reviewed quotation spans**, split by book into train / validation / test.
57
 
58
  Character offsets are **inclusive on both ends**: the span is `text[start:end+1]`.
59
 
60
+ ## Data source and preparation
61
+
62
+ **Source.**
63
 
64
  Classical Buddhist commentaries digitized with support from the Tsadra Foundation and OpenPecha, annotated in two batches: an old batch (125 books, `Quotation` layer) and a new batch (195 books, `Citation` layer). Both layers mark the same thing — one text quoting a scripture, a named teacher, or a stock phrase — so they are merged into one label.
65
 
66
+ **Cleaning**, in order:
67
 
68
  1. **Human review.** 398 books reviewed; 78 rejected whole for bad boundaries; 254 boundaries repaired in accepted books.
69
  2. **Span merging.** The new batch often splits one quotation across lines. Pieces separated only by 0–2 spaces, newlines, or a shad (and not already closed) were joined: 28,892 → 23,476 spans.
 
112
 
113
  ```python
114
  from datasets import load_dataset
115
+ ds = load_dataset("BDRC/Bo-Quotation-Dataset")
116
  ```
117
 
118
  ## Windowed configs (for mmBERT training)
 
129
  ```python
130
  from datasets import load_dataset
131
  cfg = "windowed_w8192_s4579" # or windowed_w8192_s4579_bioe
132
+ base = f"hf://datasets/BDRC/Bo-Quotation-Dataset/{cfg}"
133
  ds = load_dataset("parquet", data_files={
134
  s: f"{base}/{s}/*.parquet" for s in ["train", "validation", "test"]
135
  })
 
160
  title = {Tibetan Quotation Detection Benchmark Dataset},
161
  author = {karma689},
162
  year = {2026},
163
+ url = {https://huggingface.co/datasets/BDRC/Bo-Quotation-Dataset}
164
  }
165
  ```
166
 
 
201
 
202
  ```python
203
  from datasets import load_dataset
204
+ test = load_dataset("BDRC/Bo-Quotation-Dataset",
205
  data_files={"test": "windowed_w8192_s4579_bioe_reviewed/test/*.parquet"},
206
  revision="w2.1")["test"]
207
  ```