Datasets:
Lino-Urdaneta-Mammut
commited on
Commit
·
5af060c
1
Parent(s):
31d6b81
Update README.md
Browse files
README.md
CHANGED
@@ -25,7 +25,7 @@ The dataset contains Venezuelan and Latin-American Spanish.
|
|
25 |
|
26 |
Dataset structure features.
|
27 |
|
28 |
-
|
29 |
|
30 |
An example from the dataset:
|
31 |
|
@@ -41,12 +41,12 @@ An example from the dataset:
|
|
41 |
|
42 |
The average word token count are provided below:
|
43 |
|
44 |
-
|
45 |
|
46 |
Train: 92,431,194.
|
47 |
Test: 4,876,739 (in another file).
|
48 |
|
49 |
-
|
50 |
|
51 |
The data have several fields:
|
52 |
|
@@ -58,7 +58,7 @@ The data have several fields:
|
|
58 |
TOKENS: number of tokens (excluding punctuation marks) of SENTENCE.
|
59 |
TYPE: linguistic register of the text.
|
60 |
|
61 |
-
|
62 |
|
63 |
The mammut-corpus-venezuela dataset has 2 splits: train and test. Below are the statistics:
|
64 |
|
@@ -69,11 +69,11 @@ Test: 157,011.
|
|
69 |
|
70 |
# 6. Dataset Creation
|
71 |
|
72 |
-
|
73 |
|
74 |
The purpose of the mammut-corpus-venezuela dataset is language modeling. It can be used for pre-training a model from scratch or for fine-tuning on another pre-trained model.
|
75 |
|
76 |
-
|
77 |
|
78 |
**6.2.1 Initial Data Collection and Normalization**
|
79 |
|
@@ -89,7 +89,7 @@ Text sources: El Estímulo (website), cinco8 (website), csm-1990 (oral speaking
|
|
89 |
|
90 |
The texts come from Venezuelan Spanish speakers, subtitlers, journalists, politicians, doctors, writers, and online sellers.
|
91 |
|
92 |
-
|
93 |
|
94 |
**6.3.1 Annotation process**
|
95 |
|
@@ -99,21 +99,21 @@ At the moment the dataset does not contain any additional annotations.
|
|
99 |
|
100 |
Not applicable.
|
101 |
|
102 |
-
|
103 |
|
104 |
The data is partially anonymized. Also, there are messages from Telegram selling chats, some percentage of these messages may be fake or contain misleading or offensive language.
|
105 |
|
106 |
# 7. Considerations for Using the Data
|
107 |
|
108 |
-
|
109 |
|
110 |
The purpose of this dataset is to help the development of language modeling models (pre-training or fine-tuning) in Venezuelan Spanish.
|
111 |
|
112 |
-
|
113 |
|
114 |
Most of the content comes from political, economical and sociological opinion articles. Social biases may be present.
|
115 |
|
116 |
-
|
117 |
|
118 |
(If applicable, description of the other limitations in the data.)
|
119 |
|
@@ -121,18 +121,18 @@ Not applicable.
|
|
121 |
|
122 |
# 8. Additional Information
|
123 |
|
124 |
-
|
125 |
|
126 |
The data was originally collected by Lino Urdaneta and Miguel Riveros from Mammut.io.
|
127 |
|
128 |
-
|
129 |
|
130 |
Not applicable.
|
131 |
|
132 |
-
|
133 |
|
134 |
Not applicable.
|
135 |
|
136 |
-
|
137 |
|
138 |
Not applicable.
|
|
|
25 |
|
26 |
Dataset structure features.
|
27 |
|
28 |
+
## 5.1 Data Instances
|
29 |
|
30 |
An example from the dataset:
|
31 |
|
|
|
41 |
|
42 |
The average word token count are provided below:
|
43 |
|
44 |
+
## 5.2 Total of tokens (no spelling marks)
|
45 |
|
46 |
Train: 92,431,194.
|
47 |
Test: 4,876,739 (in another file).
|
48 |
|
49 |
+
## 5.3 Data Fields
|
50 |
|
51 |
The data have several fields:
|
52 |
|
|
|
58 |
TOKENS: number of tokens (excluding punctuation marks) of SENTENCE.
|
59 |
TYPE: linguistic register of the text.
|
60 |
|
61 |
+
## 5.4 Data Splits
|
62 |
|
63 |
The mammut-corpus-venezuela dataset has 2 splits: train and test. Below are the statistics:
|
64 |
|
|
|
69 |
|
70 |
# 6. Dataset Creation
|
71 |
|
72 |
+
## 6.1 Curation Rationale
|
73 |
|
74 |
The purpose of the mammut-corpus-venezuela dataset is language modeling. It can be used for pre-training a model from scratch or for fine-tuning on another pre-trained model.
|
75 |
|
76 |
+
## 6.2 Source Data
|
77 |
|
78 |
**6.2.1 Initial Data Collection and Normalization**
|
79 |
|
|
|
89 |
|
90 |
The texts come from Venezuelan Spanish speakers, subtitlers, journalists, politicians, doctors, writers, and online sellers.
|
91 |
|
92 |
+
# 6.3 Annotations
|
93 |
|
94 |
**6.3.1 Annotation process**
|
95 |
|
|
|
99 |
|
100 |
Not applicable.
|
101 |
|
102 |
+
## 6.4 Personal and Sensitive Information
|
103 |
|
104 |
The data is partially anonymized. Also, there are messages from Telegram selling chats, some percentage of these messages may be fake or contain misleading or offensive language.
|
105 |
|
106 |
# 7. Considerations for Using the Data
|
107 |
|
108 |
+
## 7.1 Social Impact of Dataset
|
109 |
|
110 |
The purpose of this dataset is to help the development of language modeling models (pre-training or fine-tuning) in Venezuelan Spanish.
|
111 |
|
112 |
+
## 7.2 Discussion of Biases
|
113 |
|
114 |
Most of the content comes from political, economical and sociological opinion articles. Social biases may be present.
|
115 |
|
116 |
+
## 7.3 Other Known Limitations
|
117 |
|
118 |
(If applicable, description of the other limitations in the data.)
|
119 |
|
|
|
121 |
|
122 |
# 8. Additional Information
|
123 |
|
124 |
+
## 8.1 Dataset Curators
|
125 |
|
126 |
The data was originally collected by Lino Urdaneta and Miguel Riveros from Mammut.io.
|
127 |
|
128 |
+
## 8.2 Licensing Information
|
129 |
|
130 |
Not applicable.
|
131 |
|
132 |
+
## 8.3 Citation Information
|
133 |
|
134 |
Not applicable.
|
135 |
|
136 |
+
## 8.4 Contributions
|
137 |
|
138 |
Not applicable.
|