Learning Translations via Images

visual array of images

Students

Engineering and Applied Sciences

Faculty

Associate Professor

Project Summary

Throughout my summer as a PURM mentee, I not only participated in and contributed to several innovative Natural Language Processing projects, but also gained an awareness of the lifestyles and daily routines of the professors and graduate students around me. The main project in which I was involved, titled “Learning Word Translations via Images,” was based on the idea that although some “low-resource” languages do not have many translated texts from which to build machine translation systems, the internet can contain considerable amounts of user-generated content in these languages, and this content is often associated with images. Since images are language-independent, we can use them to glean insight into the text and establish tentative word-to-word correspondences which can improve machine translation systems for these low-resource languages. My task specifically was to build a system that would read in a provided set words in both English and a target language, compile a set of images corresponding to those words (along with their neural-network-generated vector representations), and calculate the image-based similarities of these words ‘on the fly’ while visually demonstrating the computations taking place.

Even though this project was my main assignment of the summer, it was far from my only learning opportunity. Through the entire summer program, I was provided with ample materials covering topics such as vector embeddings, neural networks, and sequence-to-sequence models, and whenever something in these materials appeared too convoluted to understand, I was always able to turn to my PURM mentor or one of his graduate students for assistance. Furthermore, I attended countless meetings, workshops, and “clunches” (conference-lunches), thus acquiring a window into the life of an academic researcher: a life filled with international conferences, publications, collaborations, and seemingly intractable problems crumbling under the onslaught of mental pickaxes. Consequently, I finished the summer with both a realistic picture of life in academia and a reasonable understanding of the mysterious beast known as Artificial Intelligence.