This paper explores the possibility of unsupervised machine translation using only monolingual corpora. The authors propose a model that maps sentences from two different languages into a shared latent space, allowing the model to learn to translate without any labeled data. The model is trained to reconstruct sentences in both languages from this shared feature space. The effectiveness of the model is demonstrated on two widely used datasets (Multi30k and WMT) for English-French translation, achieving BLEU scores of 32.8 and 15.1, respectively, without using any parallel sentences during training. The key idea is to build a common latent space between the two languages and learn to translate by reconstructing sentences in both domains using denoising auto-encoders and cross-domain losses. The model is initialized with a naive unsupervised translation model based on word-by-word translation using a bilingual lexicon derived from monolingual data. The paper also discusses related work and provides experimental results to validate the effectiveness of the proposed approach.This paper explores the possibility of unsupervised machine translation using only monolingual corpora. The authors propose a model that maps sentences from two different languages into a shared latent space, allowing the model to learn to translate without any labeled data. The model is trained to reconstruct sentences in both languages from this shared feature space. The effectiveness of the model is demonstrated on two widely used datasets (Multi30k and WMT) for English-French translation, achieving BLEU scores of 32.8 and 15.1, respectively, without using any parallel sentences during training. The key idea is to build a common latent space between the two languages and learn to translate by reconstructing sentences in both domains using denoising auto-encoders and cross-domain losses. The model is initialized with a naive unsupervised translation model based on word-by-word translation using a bilingual lexicon derived from monolingual data. The paper also discusses related work and provides experimental results to validate the effectiveness of the proposed approach.