Combining Word and Character Embeddings for Arabic Chatbots

The Arabic language has a rich morphology and structure with a diverse vocabulary and rarely used words. Consequently, most Arabic Natural Language Processing (NLP) tasks could benefit from embedding models that do not assign a distinct vector to each unique word in the used vocabulary but instead focus on the internal structure of words. The semantic meaning of a word is related to the meaning of its composing characters which contain rich internal information. In this paper, we propose a new embedding model using two levels of granularity; words and characters. Moreover, we describe the details of generating an Arabic word embeddings using Word2Vec and FastText models. Furthermore, a Deep Learning (DL) architecture will be applied to the top of the word-character embeddings. Experimental results show that the proposed scheme outperforms the state-of-the-art methods proposed for Arabic chatbots.