Adapting Google Translate using Dictionary and Word Embedding for Arabic-Indonesian Cross-lingual Information Retrieval

The translation has an essential role in Cross-lingual Information Retrieval. Translation using a dictionary is reliable even though it has a limited vocabulary. Translation using google translate, in some cases, using different words used in document target words. The translation process causes word translation to be less accurate to get relevant documents. In this paper, we proposed a new translation approach by adapting google translate using a dictionary and word embedding in Arabic-Indonesian Cross-lingual Information Retrieval. The dictionary is the primary resource used for translation improved by Levenshtein distance and FastText for finding the correct word translation. Google translate is used to complete translation when the word does not exist in the dictionary resource. The proposed method archive a BLEU score of 0.47. This score is higher than the other comparison resource score. The proposed method successfully improves the translated query to retrieve more relevant documents in cross-lingual information retrieval based on this implementation.