Commit
·
5b580f2
1
Parent(s):
8143770
Update README.md
Browse files
README.md
CHANGED
@@ -16,6 +16,61 @@ dataset_info:
|
|
16 |
download_size: 3052451421
|
17 |
dataset_size: 4637918210
|
18 |
---
|
19 |
-
# Dataset Card for "constrained_language"
|
20 |
|
21 |
-
[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
download_size: 3052451421
|
17 |
dataset_size: 4637918210
|
18 |
---
|
|
|
19 |
|
20 |
+
# Dataset Card for [Dataset Name]
|
21 |
+
|
22 |
+
## Table of Contents
|
23 |
+
- [Dataset Description](#dataset-description)
|
24 |
+
- [Dataset Summary](#dataset-summary)
|
25 |
+
- [Languages](#languages)
|
26 |
+
- [Dataset Structure](#dataset-structure)
|
27 |
+
- [Data Instances](#data-instances)
|
28 |
+
- [Citation Information](#additional-information)
|
29 |
+
- [Citation Information](#citation-information)
|
30 |
+
|
31 |
+
|
32 |
+
## Dataset Description
|
33 |
+
|
34 |
+
- **Paper: https://arxiv.org/abs/2305.17266**
|
35 |
+
- **Point of Contact: [email protected]**
|
36 |
+
|
37 |
+
### Dataset Summary
|
38 |
+
|
39 |
+
This dataset is one of the two datasets published by \textit{Honey, I Shrunk the Language: Language Model Behavior at Reduced Scale}.
|
40 |
+
The dataset available at this link is the pre-training data constrained by vocabulary. The other published data i.e. the pre-training data that is not constrained by vocabulary is available at https://huggingface.co/datasets/text-machine-lab/unconstrained_language.
|
41 |
+
|
42 |
+
The vocabulary used for curating the data is constructed from the AOChildes corpus (https://www.sciencedirect.com/science/article/abs/pii/S0079742121000256). The AOChildes corpus consists of transcripts of child-directed speech. Hence, the vocabulary constructed from AOChildes corpus consists of words spoken or heard by children of age six years or younger.
|
43 |
+
|
44 |
+
The vocabulary is then used to filter the widely used text corpora,
|
45 |
+
- C4: https://arxiv.org/abs/1910.10683,
|
46 |
+
- BookCorpus: https://ieeexplore.ieee.org/document/7410368,
|
47 |
+
- Wikipedia: https://huggingface.co/datasets/wikipedia,
|
48 |
+
- Simplified-Wikipedia: https://simple.wikipedia.org/wiki/Main_Page,
|
49 |
+
- Childrens' Book Test Corpus: https://arxiv.org/abs/1511.02301
|
50 |
+
|
51 |
+
From above corpora, only those spans are included in the that that constains words only from the predefined vocabulary. The dataset includes 44 million sentences (~6 million sequences of ~128 tokens) and 3 million contiguous spans with approximately 128 tokens.
|
52 |
+
|
53 |
+
### Languages
|
54 |
+
|
55 |
+
The dataset contains English language only.
|
56 |
+
|
57 |
+
## Dataset Structure
|
58 |
+
|
59 |
+
The dataset is available in the arrow dataset format with three splits namely, train, validation and test.
|
60 |
+
|
61 |
+
### Data Instances
|
62 |
+
|
63 |
+
Every data instance has only one key "Text" that included a text span of approximately 128 tokens.
|
64 |
+
|
65 |
+
### Citation Information
|
66 |
+
|
67 |
+
```
|
68 |
+
@article{deshpande2023honey,
|
69 |
+
title={Honey, I Shrunk the Language: Language Model Behavior at Reduced Scale},
|
70 |
+
author={Deshpande, Vijeta and Pechi, Dan and Thatte, Shree and Lialin, Vladislav and Rumshisky, Anna},
|
71 |
+
journal={arXiv preprint arXiv:2305.17266},
|
72 |
+
year={2023}
|
73 |
+
}
|
74 |
+
```
|
75 |
+
|
76 |
+
|