To that end, we are developing conversational interfaces for multimodal search and question answering, with long term sights on applications that can learn, teach, explain and fluidly interact with users in their everyday lives.
Adversarially Augmented Training
Conversational search is messy. Misspellings, incorrect grammar and ambiguous references are all part of a day's work for conversational search engines, and the language models that power them must be robust enough to handle all types of nuances.
Multimodal Question Answering
Useful information is buried in documents of all types, from PDFs to images to text files. In order to truly leverage all of the hidden knowledge in these files, conversational systems need to understand each modality in the context of the full document collection and patch together the puzzle pieces to deliver new insights to challenging questions.
Chat interfaces have a long way to go before they start sounding like HAL 9000. At Searchable, we are building conversational agents and training environments that can aid in information retrieval with the fluidity and precision of a real-world assistant.