Collaborative Language Engine (CLE) - overview
Team environments in the 21st century are increasingly distributed and globalized. Collaboration solutions have become essential to team productivity. However, information overload resulting from such software is all too common. We use multiple tools such as instant chats and emails to communicate with our colleagues, and receive numerous messages that may not be important for our tasks at hand. This vast amount of irrelevant information occupies our cognitive attention, away from the (few) truly important messages. Furthermore, the productivity-boosting potential of collaboration solutions has yet to be fully unearthed. One key reason is that conversation messages are simply treated as dialogs, and often disconnected from elements that matter to the productivity of users: preferences, goals and actual work contexts.
Broadly, CLE is about conversations that matter to you. It is a cross-disciplinary research program around deep understanding of conversational meaning and contexts, as well as leveraging such an understanding to improve productivity in collaboration. This is achieved via developing world-class solutions at the intersection of AI, linguistics, user experience and privacy. The following themes are being investigated in CLE:
- Conversational message classification using cutting-edge natural language processing (NLP) and machine learning techniques;
- Identifying conversational flows;
- Context-awareness research connecting message classifications to wider work-related contexts;
- Creating user experiences that improve user engagement and productivity around collaborative conversations;
- Privacy-preserving machine learning.
CLE existed in production as a core NLP engine behind IBM Watson Workspace. It is a state-of-the-art, differentiating capability that offered IBM Watson Workspace a competitive edge. CLE in production highlightd two important types of messages 'Requests' (Actions in the screen shot above), things that are asked of a user; and 'Commitments', things that a user has promised to do. Both of these statements are refered to as 'Actionable Statements'. It could also identify questions.
For questions, please contact lead scientist Rui Zhang.