Researchers explored how to improve machine translation by extracting translations from the web and enhancing language models. They focused on using microblog platforms for translation data and developed new methods for language identification. They also worked on improving language models for tasks like reranking and morphological analysis. By analyzing syntactic features and using a novel algorithm for exact inference, they were able to better utilize existing multilingual corpora. The goal was to enhance translation quality by extracting relevant training data and improving language models.