Continuous and Interactive Language Learning and Grounding

Many task-oriented chatbots and virtual assistants like Siri, Alexa, and Google Assistant are built as Natural Language (command) Interfaces (NLIs) that allow users to issue natural language (NL) commands to be mapped to some actions for execution in the underlying application in order to accomplish some tasks intended by the users. A fundamental feature of such systems is the ability to understand users’ language and ground them to intended actions (often in symbolic form). Due to their diverse and wide-spread real-world applications, such NLI systems have driven research in language understanding, grounding and human-robot interactions over the years. This chapter discusses the scope for continual and interactive language learning in the context of NLIs and introduces some of the representative works along this direction.