Language-aware applications such as Google Translate and Siri have become useful tools of everyday lives. Despite their significant advances, they are challenged by nuances of language in the wild (e.g., idioms). In the first part of the talk, I will describe an unsupervised computational formulation to detect the figurative sense in a context-specific way. The test is computationally simple, relying on no external resources other than a set of trained word vectors. The key insight is connecting how the meaning of a phrase relates to the context it is in, using a fundamental geometric property. This leads to new state of the art results on idiomaticity detection on a variety of datasets in multiple languages. This approach also provides a common computational umbrella to study linguistically different natural language phenomena (idiomaticity, sarcasm, metaphor). Cyberspace (e.g., social platforms) and healthcare contexts pose challenges to human sensibility via language nuances. In the second part of the talk, I will discuss concrete engineering solutions to correct malicious spelling errors and to simplify patient health data; these are collaborative efforts with psychologists and health professionals.

Suma Bhat is Assistant Professor in ECE at Illinois. Her core research focuses on computational linguistics and natural language processing. She actively collaborates on finding natural language processing solutions to challenges involving textual data in healthcare, education and cyberspace control.

Host by: Heng Ji