Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -79,7 +79,7 @@ This still growing experimental project/dataset is not connected, funded or orga
|
|
| 79 |
|
| 80 |
## Overview
|
| 81 |
|
| 82 |
-
At the moment the Leipzig Corpora Collention has many raw corpora transparently documented and available for downloads since 1995 till 2024 – this [german only news collection starts from 2012 till 2024](https://wortschatz.uni-leipzig.de/de/download/German#corpora_type_news) – I go back in time from time to time to add the next year.
|
| 83 |
|
| 84 |
Aside the german part of wikipedia contents there are still not really much datasets for actual content under a free license. Backtranslating from english to a different language with translation models getting better troughout the years, is not really new. This dataset collection is an experiment using german only sentences from the LCC news section. The sentences in this collection are unfiltered but scored after a translation from german to english and a following backtranslation to german again. To get a wider variance of possible synonyms in the backtranslation process the parameters have been 'widened' using ctranslate-based [quickmt - Neural Machine Translation Library](https://github.com/quickmt/quickmt).
|
| 85 |
|
|
|
|
| 79 |
|
| 80 |
## Overview
|
| 81 |
|
| 82 |
+
At the moment the Leipzig Corpora Collention has many raw monolingual corpora transparently documented and available for downloads since 1995 till 2024 – this [german only news collection starts from 2012 till 2024](https://wortschatz.uni-leipzig.de/de/download/German#corpora_type_news) – I go back in time from time to time to add the next year.
|
| 83 |
|
| 84 |
Aside the german part of wikipedia contents there are still not really much datasets for actual content under a free license. Backtranslating from english to a different language with translation models getting better troughout the years, is not really new. This dataset collection is an experiment using german only sentences from the LCC news section. The sentences in this collection are unfiltered but scored after a translation from german to english and a following backtranslation to german again. To get a wider variance of possible synonyms in the backtranslation process the parameters have been 'widened' using ctranslate-based [quickmt - Neural Machine Translation Library](https://github.com/quickmt/quickmt).
|
| 85 |
|