Datasets:

projecte-aina
/

sts-ca

@@ -47,7 +47,7 @@ The dataset is in Catalan (`ca-CA`).
 ### Data Instances
-Follows SemEval challenges (https://www.aclweb.org/anthology/S13-1004.pdf).
 * index (int)
 * id (str): Unique ID assigned to the sentence pair.
 * sentence 1 (str): First sentence of the pair.
@@ -82,7 +82,7 @@ This dataset follows [SemEval](https://www.aclweb.org/anthology/S13-1004.pdf) ch
 ### Methodology
-Random sentences were extracted from 3 Catalan corpus: ACN, Oscar and Wikipedia, and we generated candidate pairs using a combination of metrics from Doc2Vec, Jaccard and a BERT-like model (“distiluse-base-multilingual-cased-v2”, [link](https://huggingface.co/distilbert-base-multilingual-cased)). Finally, we  manually reviewed the generated pairs to reject non-relevant pairs (identical or ungrammatical sentences, etc.) before providing them to the annotation team.
 The average of the four annotations was selected as a “ground truth” for each sentence pair, except when an annotator diverged in more than one unit from the average. In these cases, we discarded the divergent annotation and recalculated the average without it. We also discarded 45 sentence pairs because the annotators disagreed too much.
@@ -98,7 +98,7 @@ The source data are scraped sentences from the [Catalan Textual Corpus](https://
 #### Who are the source language producers?
-The Catalan Textual Corpus is a 1760-million-token web corpus of Catalan built from several sources: existing corpus such as DOGC, CaWac (non-dedup version), Oscar (unshuffled version), Open Subtitles, Catalan Wikipedia; and three brand new crawlings: the Catalan General Crawling, obtained by crawling the 500 most popular .cat and .ad domains; the Catalan Government Crawling, obtained by crawling the .gencat domain and subdomains, belonging to the Catalan Government; and the ACN corpus with 220k news items from March 2015 until October 2020, crawled from the Catalan News Agency.
 ### Annotations

 ### Data Instances
+Follows [SemEval challenges](https://www.aclweb.org/anthology/S13-1004.pdf):
 * index (int)
 * id (str): Unique ID assigned to the sentence pair.
 * sentence 1 (str): First sentence of the pair.
 ### Methodology
+Random sentences were extracted from 3 Catalan corpus: ACN, Oscar and Wikipedia, and we generated candidate pairs using a combination of metrics from Doc2Vec, Jaccard and a BERT-like model (“[distiluse-base-multilingual-cased-v2](https://huggingface.co/distilbert-base-multilingual-cased)”). Finally, we  manually reviewed the generated pairs to reject non-relevant pairs (identical or ungrammatical sentences, etc.) before providing them to the annotation team.
 The average of the four annotations was selected as a “ground truth” for each sentence pair, except when an annotator diverged in more than one unit from the average. In these cases, we discarded the divergent annotation and recalculated the average without it. We also discarded 45 sentence pairs because the annotators disagreed too much.
 #### Who are the source language producers?
+The [Catalan Textual Corpus](https://zenodo.org/record/4519349#.Ys_0PexBzOs) is a 1760-million-token web corpus of Catalan built from several sources: existing corpus such as DOGC, CaWac (non-deduplicated version), Oscar (unshuffled version), Open Subtitles, Catalan Wikipedia; and three brand new crawlings: the Catalan General Crawling, obtained by crawling the 500 most popular .cat and .ad domains; the Catalan Government Crawling, obtained by crawling the .gencat domain and subdomains, belonging to the Catalan Government; and the ACN corpus with 220k news items from March 2015 until October 2020, crawled from the Catalan News Agency.
 ### Annotations