GhezaeGoitom/Tigringna_and_Geez_Word_Embeddings

This embeddings were trained using fastText with the skip-gram model. For Tigrinya, it is trained by collecting textual data from Eritrean newspapers and magazines, while for Geez, it was derived from biblical and other reference books. # Tigrinya and Geez Word Embeddings # Intro This embeddings were trained using fastText with the skip-gram model. For Tigrinya, this embedding was created by collecting textual data from Eritrean newspapers and magazines, while for Geez, it was derived from biblical and other reference books. ## Tigrinya The Tigrinya embeddings were trained using data extracted from Eritrean newspapers, particularly from ”Haddas Ertra”. This PDF repository encompassed approximately 600 articles spanning from January 1, 2023, to June 1, 2023. The selection of text encompassed various domains or topics covered by Haddas Ertra. Furthermore, we also acquired an additional, publicly accessible corpus tagged with part-of-speech (PoS) for the Tigrinya language, known as "The Nagaoka Tigrinya Corpus 1.0 (NTC 1.0)". This corpus was compiled by collecting news articles from the Eritrean newspaper "Haddas Ertra". The creator of the corpus reports that it comprises approximately 100 articles that were published between March 2013 and December 2013, containing a total of 72,080 tokens. On average, each sentence consists of approximately 15 words. Additionally, we obtained data from the Eritrean magazine called "Menesey Magazine", which spans a period of two years (2022-2023) and encompasses diverse topics. ## Geez As for Geez, the training data was sourced from various historical materials. Notably, we utilized the publicly accessible dataset titled "TraCES Corpus of the Classical Ethiopic Language (Ge'ez)" developed by the University of Hamburg. This dataset incorporates textual content from 11 chapters of the ancient Bible. Additionally, we gathered further historical resources. These resources are classified into three distinct domains, encompassing one educational book, one historical book, and an additional five books from the religious domain.