rollback changes

README.md

@@ -56,14 +56,19 @@ task_ids:
 
 ### Dataset Summary
 
-The `tldr_news` dataset was constructed by collecting a daily tech newsletter (
-available [here](https://tldr.tech/newsletter)). Then for every piece of news, the "headline" and its corresponding "
-content" were collected.
+The `tldr_news` dataset was constructed by collecting a daily tech newsletter (available
+[here](https://tldr.tech/newsletter)). Then, for every piece of news, the `headline` and its corresponding `
+content` were extracted.
+
 Such a dataset can be used to train a model to generate a headline from a input piece of text.
 
 ### Supported Tasks and Leaderboards
 
-There is no official supported tasks nor leaderboard for this dataset.
+There is no official supported tasks nor leaderboard for this dataset. However, it could be used for the following
+tasks:
+
+- summarization
+- headline generation
 
 ### Languages
 
@@ -90,22 +95,31 @@ An example is as follows:
 
 ### Data Splits
 
-[Needs More Information]
+- `all`: all existing daily newsletters available [here](https://tldr.tech/newsletter).
 
 ## Dataset Creation
 
 ### Curation Rationale
 
 This dataset was obtained by scrapping the collecting all the existing newsletter
-available [here](https://tldr.tech/newsletter). Every single newsletter was then processed to extract all the different
-pieces of news. Then for every collected piece of news the headline and the content were extracted.
+available [here](https://tldr.tech/newsletter).
+
+Every single newsletter was then processed to extract all the different pieces of news. Then for every collected piece
+of news the headline and the news content were extracted.
 
 ### Source Data
 
 #### Initial Data Collection and Normalization
 
-The dataset was has been collected from https://tldr.tech/newsletter. The headline and content have not been
-post-processed nor normalized.
+The dataset was has been collected from https://tldr.tech/newsletter.
+
+In order to clean up the samples and to construct a dataset better suited for headline generation we have applied a
+couple of normalization steps:
+
+1. The headlines initially contain an estimated read time in parentheses; we stripped this information from the
+   headline.
+2. Some headlines are just a technology/repo/software name. We filtered out such samples if this name was not mention in
+   the news content.
 
 #### Who are the source language producers?
 
@@ -115,12 +129,12 @@ The people (or person) behind the https://tldr.tech/ newsletter.
 
 #### Annotation process
 
-[Needs More Information]
+Disclaimers: The dataset was generated from a daily newsletter. The author had no intention for those newsletters to be
+used as such.
 
 #### Who are the annotators?
 
-Disclaimers: The dataset was generated from a daily newsletter. The author had no intention for those newsletters to be
-used as such.
+The newsletters were written by the people behind *TLDR tech*.
 
 ### Personal and Sensitive Information
 
@@ -134,17 +148,17 @@ used as such.
 
 ### Discussion of Biases
 
-[Needs More Information]
+This dataset only contains tech news. A model trained on such a dataset might not be able to generalize to other domain.
 
 ### Other Known Limitations
 
-This dataset only contains tech news.
+[Needs More Information]
 
 ## Additional Information
 
 ### Dataset Curators
 
-The dataset was obtained by scrapping this website: https://tldr.tech/newsletter
+The dataset was obtained by collecting newsletters from this website: https://tldr.tech/newsletter
 
 ### Contributions