Microsoft Icecaps: An Open-Source Toolkit for Conversation Modeling

Association for Computational Linguistics 2019 |

Organized by Microsoft

The Intelligent Conversation Engine: Code and Pre-trained Systems (Microsoft Icecaps) is an upcoming open-source natural language processing repository. Icecaps wraps TensorFlow functionality in a modular component-based architecture, presenting an intuitive and flexible paradigm for constructing sophisticated learning setups. Capabilities include multi-task learning between models with shared parameters, upgraded language model decoding features, a range of built-in architectures, and a user-friendly data processing pipeline. The system is targeted toward conversational tasks, exploring diverse response generation, coherence, and knowledge grounding. Icecaps also provides pre-trained conversational models that can be either used directly or loaded for fine-tuning or bootstrapping other models; these models power an online demo of our framework. (These pre-trained models and demo will be included in a future update of Icecaps.)