Modeling Lexically Divergent Paraphrases in Twitter and Beyond - 2015

Explore the challenges of natural language processing and the power of paraphrases in this 1-hour lecture by Wei Xu from the University of Pennsylvania. Discover a novel multi-instance learning model that captures diverse paraphrases from Twitter data, including synonyms, acronyms, misspellings, and colloquialisms. Learn how paraphrases can be applied to adapt statistical machine translation techniques for text-to-text generation tasks like text simplification and stylistic rewriting. Examine the application of similar models for distantly supervised information extraction using large knowledge bases. Gain insights into cutting-edge research on paraphrases, social media, and information extraction from an expert in the field.