Datasets:

mammut
/

mammut-corpus-venezuela

@@ -23,7 +23,9 @@ The dataset contains Venezuelan and Latin-American Spanish.
 # 5. Dataset Structure
-## 5.1 Data Instances
 An example from the dataset:
@@ -39,12 +41,12 @@ An example from the dataset:
 The average word token count are provided below:
-## 5.2 Total of tokens (no spelling marks)**
 Train: 92,431,194.
 Test: 4,876,739 (in another file).
-## 5.3 Data Fields
 The data have several fields:
@@ -56,7 +58,7 @@ The data have several fields:
   TOKENS: number of tokens (excluding punctuation marks) of SENTENCE.
   TYPE: linguistic register of the text.
-## 5.4 Data Splits
 The mammut-corpus-venezuela dataset has 2 splits: train and test. Below are the statistics:
@@ -67,13 +69,13 @@ Test: 157,011.
 # 6. Dataset Creation
-## 6.1 Curation Rationale
 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.
-## 6.2 Source Data
-### 6.2.1 Initial Data Collection and Normalization
 The data consists of opinion articles and text messages. It was collected by a process of web scraping from different portals, downloading of Telegram group chats’ history and selecting of Venezuelan and Latin-American Spanish corpus available online.
@@ -83,11 +85,11 @@ An arrow parquet file was created.
 Text sources: El Estímulo (website), cinco8 (website), csm_1990 (oral speaking corpus), "El atajo más largo" (blog), El Pitazo (website), La Patilla (website), Venezuelan movies subtitles, Preseea Mérida (oral speaking corpus), Prodavinci (website), Runrunes (website), and Telegram group chats.
-### 6.2.2 Who are the source language producers?
 The texts come from Venezuelan Spanish speakers, subtitlers, journalists, politicians, doctors, writers, and online sellers.
-## 6.3 Annotations
 ### 6.3.1 Annotation process
@@ -97,39 +99,40 @@ At the moment the dataset does not contain any additional annotations.
 Not applicable.
-## 6.4 Personal and Sensitive Information
 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.
 # 7. Considerations for Using the Data
-## 7. 1Social Impact of Dataset
 The purpose of this dataset is to help the development of language modeling models (pre-training or fine-tuning) in Venezuelan Spanish.
-## 7.2 Discussion of Biases
 Most of the content comes from political, economical and sociological opinion articles. Social biases may be present.
-## 7.3 Other Known Limitations
 (If applicable, description of the other limitations in the data.)
 Not applicable.
 # 8. Additional Information
-## 8.1 Dataset Curators
 The data was originally collected by Lino Urdaneta and Miguel Riveros from Mammut.io.
-## 8.2 Licensing Information
 Not applicable.
-## 8.3 Citation Information
 Not applicable.
-## 8.4 Contributions
 Not applicable.

 # 5. Dataset Structure
+Dataset structure features.
+**5.1 Data Instances**
 An example from the dataset:
 The average word token count are provided below:
+**5.2 Total of tokens (no spelling marks)**
 Train: 92,431,194.
 Test: 4,876,739 (in another file).
+**5.3 Data Fields**
 The data have several fields:
   TOKENS: number of tokens (excluding punctuation marks) of SENTENCE.
   TYPE: linguistic register of the text.
+**5.4 Data Splits**
 The mammut-corpus-venezuela dataset has 2 splits: train and test. Below are the statistics:
 # 6. Dataset Creation
+**6.1 Curation Rationale**
 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.
+**6.2 Source Data**
+**6.2.1 Initial Data Collection and Normalization**
 The data consists of opinion articles and text messages. It was collected by a process of web scraping from different portals, downloading of Telegram group chats’ history and selecting of Venezuelan and Latin-American Spanish corpus available online.
 Text sources: El Estímulo (website), cinco8 (website), csm_1990 (oral speaking corpus), "El atajo más largo" (blog), El Pitazo (website), La Patilla (website), Venezuelan movies subtitles, Preseea Mérida (oral speaking corpus), Prodavinci (website), Runrunes (website), and Telegram group chats.
+**6.2.2 Who are the source language producers?**
 The texts come from Venezuelan Spanish speakers, subtitlers, journalists, politicians, doctors, writers, and online sellers.
+**6.3 Annotations**
 ### 6.3.1 Annotation process
 Not applicable.
+**6.4 Personal and Sensitive Information**
 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.
 # 7. Considerations for Using the Data
+**7.1 Social Impact of Dataset**
 The purpose of this dataset is to help the development of language modeling models (pre-training or fine-tuning) in Venezuelan Spanish.
+**7.2 Discussion of Biases**
 Most of the content comes from political, economical and sociological opinion articles. Social biases may be present.
+**7.3 Other Known Limitations**
 (If applicable, description of the other limitations in the data.)
 Not applicable.
 # 8. Additional Information
+**8.1 Dataset Curators**
 The data was originally collected by Lino Urdaneta and Miguel Riveros from Mammut.io.
+**8.2 Licensing Information**
 Not applicable.
+**8.3 Citation Information**
 Not applicable.
+**8.4 Contributions**
 Not applicable.