The world of social media has undergone huge evolution during the last few years. With the spread of social media and online forums, individual users actively participate in the generation of online content in different languages from all over the world. Sharing of online content has become much easier than before with the advent of popular websites such as Twitter, Facebook etc. Such content is referred to as ‘User-Generated Content’ (UGC). Some examples of UGC are user reviews, customer feedback, tweets etc. In general, UGC is informal and noisy in terms of linguistic norms. Such noise does not create significant problems for human to understand the content, but it can pose challenges for several natural language processing applications such as parsing, sentiment analysis, machine translation (MT), etc. An additional challenge for MT is sparseness of bilingual (translated) parallel UGC corpora. In this research, we explore the general issues in MT of UGC and set some research goals from our findings. One of our main goals is to exploit comparable corpora in order to extract parallel or semantically similar sentences. To accomplish this task, we design a document alignment system to extract semantically similar bilingual document pairs using the bilingual comparable corpora. We then apply strategies to extract parallel or semantically similar sentences from comparable corpora by transforming the document alignment system into a sentence alignment system. We seek to improve the quality of parallel data extraction for UGC translation and assemble the extracted data with the existing human translated resources. Another objective of this research is to demonstrate the usefulness of MT-based sentiment analysis. However, when using openly available systems such as Google Translate, the translation process may alter the sentiment in the target language. To cope with this phenomenon, we instead build fine-grained sentiment translation models that focus on sentiment preservation in the target language during translation.