Resources

Baliabideak

In this section you can access the datasets and resources we have created. In addition to these contents, you can also access the software, neural models and evaluation data generated by our centre:

Codes and respositories

Neural language models and datasets

Opinion Mining - Sentiment Analysis

ElhPolar_es

Spanish polarity lexicon.

ElhPolar_eu

Basque polarity lexicon.

Basque Opinon Dataset

Polarity annotated Basque sentences.  

BEC2016 opinion dataset

Basque regional election campaign 2016 opinion dataset - BEC2016. 25.000 Tweets with entity level polarity annotations (pos|neg).  

Behagunea Opinion dataset

Tweet collection about the DSS2016 Cultural capital project. Tweets annotated with polarity at message level (pos|neg|neu) in Basque (3000) and Spanish (4754).  

EliXa polarity classification models (EliXa 1.0.x)

Models for polarity classification, trained over cultural domain (Behagunea) tweets. Previous versions: v 0.9.x

EliXa resources (EliXa 1.0.x <=)

language specific resources: polarity lexicons and other resources for text normalization. We currently provide such resources for 4 languages; Basque (eu), Spanish (es), English (en) and French (fr). Also includes pos tagging models for ixa-pipe-pos tool. Previous versions: v 0.9.x ( Ixa-pipes pos models not  included)

Ixa-Pipes models for EliXa 0.9.x

Ixa-Pipes models used for lemmatization and POS tagging (1.5.0) by EliXa 0.9.x as default models.  

Corpora

Basque-English Parallel corpus

Basque-English parallel corpus automatically gathered using the PaCo2 tool.  

Basque-Spanish Parallel corpus

Basque-Spanish parallel corpus automatically gathered using the PaCo2 tool. It contains 640K segments.  

Elhuyar web corpus

Corpus of 186M tokens in Basque. Automatically crawled and cleaned from the Web. Ref: Leturia, I. 2014. The Web as a Corpus of Basque. PhD Thesis. Faculty of Informatics, UPV/EHU. Donostia.  

Conversational assistants

FMTODelh dataset

Basque version of the Facebook Multilingual Task Oriented Dataset (López de Lacalle et al., 2020). Train and Dev sets have been translated using NMT. Test set has been manually translated.  

SNIPSeu dataset

SNIPS Dataset (Coucke A. et al., 2018) test set manually translated for Basque (López de Lacalle et al., 2021)  

Dokumentuen Sailkapena

BHTC dataset

Basque Headlines Document Classification (BHTC) dataset. Collection containing 12,403 headlines extracted from the weekly newspaper Argia with topic annotations. Used for document classification task (Agerri et al., 2020).  

Grammatical Error Correction

GEC-elh-eu dataset

Grammatical Error Correction (GEC) dataset for Basque. 9 million synthetic sentence pairs (incorrect - correct) as train dataset. For evaluation synthethic examples (6,000) and manual revised examples (672) are provided. If you use it, cite (Beloki et al., 2020) paper.