The success of current parsing method crucially depends on the availability of supervised data. However, for many languages and domains such annotations are not readily available and are prohibitively expensive to collect. We will investigate novel tensor-based methods for learning sentence representations that enable to train parsers from a few annotated sentences and large amounts of annotated data.

I worked with the Senseable City Lab on the Local Warming project to create a dynamically controlled localized heating system. I worked at Google to improve the user interface for the Knowledge Carousel.