Below you find links to Spanish word embeddings computed with different methods and from different corpora. Whenever it is possible, a description of the parameters used to compute the embeddings is included, together with simple statistics of the vectors, vocabulary, and description of the corpus from which the embeddings were computed. Direct links to the embeddings are provided, so please refer to the original sources for proper citation (also see References). An example of the use of some of these embeddings can be found here or in this tutorial (both in Spanish).
Summary (and links) for the embeddings in this page:
| Corpus | Size | Algorithm | #vectors | vec-dim | Credits | |
|---|---|---|---|---|---|---|
| 1 | Spanish Unannotated Corpora | 2.6B | FastText | 1,313,423 | 300 | José Cañete |
| 2 | Spanish Billion Word Corpus | 1.4B | FastText | 855,380 | 300 | Jorge Pérez |
| 3 | Spanish Billion Word Corpus | 1.4B | Glove | 855,380 | 300 | Jorge Pérez |
| 4 | Spanish Billion Word Corpus | 1.4B | Word2Vec | 1,000,653 | 300 | Cristian Cardellino |
| 5 | Spanish Wikipedia | ??? | FastText | 985,667 | 300 | FastText team |
Links to the embeddings (#dimensions=300, #vectors=1,313,423):
- Vector format (.vec) (3.4 GB)
- Binary format (.bin) (5.6 GB)
More vectors with different dimensiones (10, 30, 100, and 300) can be found here
- Implementation: FastText with Skipgram
- Parameters:
- min subword-ngram = 3
- max subword-ngram = 6
- minCount = 5
- epochs = 20
- dim = 300
- all other parameters set as default
- Spanish Unannotated Corpora
- Corpus Size: 3 billion words
- Post processing: Explained in Embeddings and Corpora repos, that include tokenization, lowercase, removed listings and urls.
Links to the embeddings (#dimensions=300, #vectors=855,380):
- Vector format (.vec.gz) (802 MB)
- Binary format (.bin) (4.2 GB)
- Implementation: FastText with Skipgram
- Parameters:
- min subword-ngram = 3
- max subword-ngram = 6
- minCount = 5
- epochs = 20
- dim = 300
- all other parameters set as default
- Spanish Billion Word Corpus
- Corpus Size: 1.4 billion words
- Post processing: Besides the post processing of the raw corpus explained in the SBWCE page that included deletion of punctuation, numbers, etc., the following processing was applied:
- Words were converted to lower case letters
- Every sequence of the 'DIGITO' keyword was replaced by (a single) '0'
- All words of more than 3 characteres plus a '0' were ommitted (example: 'padre0')
Links to the embeddings (#dimensions=300, #vectors=855,380):
- Vector format (.vec.gz) (906 MB)
- Binary format (.bin) (3.9 GB)
- Implementation: GloVe
- Parameters:
- vector-size = 300
- iter = 25
- min-count = 5
- all other parameters set as default
- Spanish Billion Word Corpus (see above)
Links to the embeddings (#dimensions=300, #vectors=1,000,653)
- Implementation: Word2Vec with Skipgram by GenSim
- Parameters: For details on parameters please refer to the SBWCE page
- Spanish Billion Word Corpus
- Corpus Size: 1.4 billion words
Links to the embeddings (#dimensions=300, #vectors=985,667):
- Vector format (.vec) (2.4 GB)
- Binary plus vector format (.zip) (5.4 GB)
- Implementation: FastText with Skipgram
- Parameters: FastText default parameters
- FastText embeddings from SUC: Word embeddings were computed by José Cañete at BotCenter. You can use these vectors as you wish under the MIT license. Please refer to BotCenter Embeddings repo for further discussion. You may also want to cite the FastText paper Enriching Word Vectors with Subword Information.
- FastText embeddings from SBWC: Word embeddings were computed by Jorge Pérez. You can use these vectors as you wish under the CC-BY-4.0 license. You may also want to cite the FastText paper Enriching Word Vectors with Subword Information and the Spanish Billion Word Corpus project.
- GloVe embeddings from SBWC: Word embeddings were computed by Jorge Pérez. You can use these vectors as you wish under the CC-BY-4.0 license. You may also want to cite the GloVe paper GloVe: Global Vectors for Word Representation and the Spanish Billion Word Corpus project.
- FastText embeddings from Spanish Wikipedia: Word embeddings were computed by FastText team. Please refer to FastText Pre-trained Vectors page if you want to use these vectors.
- Word2Vec embeddings from SBWC: Word embeddings were computed by Cristian Cardellino. Please refer to the SBWCE page if you want to use these vectors.