Spoken Language Understanding

Our research in this area focuses on recognizing and understanding user speech in languages with little data, such as code-mixing speech in Hindi-English or non-standard Chinese. We are also working on task-oriented dialog systems where external knowledge bases are integrated into an end2end deep learning framework. A current focus is to render such systems "empathetic".

Natural Language Processing

We currently focus on using deep learning methods to solve NLP problems such as bilingual lexicon extraction, abusive and toxic language detection, fake news detection etc. One of our research highlights is the joint learning of audio, speech and language, facial signals for different applications such as spoken language summarization, speech translation, and dialog systems.

Affect Recognition

We are interested in the recognition of emotion, mood, sentiment and personality from mutli-channel inputs such as speech, language, facial experession and gesture. We also analyze music audio signals to enable efficient retrieval of tens of millions of songs. We apply emotion and sentiment analysis to social media, the financial domain and creative arts. We explore emotional embeddings of words for enhanced natural langauge understanding. We propose cross-cultural embeddings for personality recognition.

Big Data Analytics

We use machine learning and deep learning approaches to discover patterns from large amounts of online data to predict user tastes and behavior, from financial data to predict market sentiment and behavior and from medical data to provide assistance to doctors.