An enriched experience with a system that answers general questions

It is hard for machines to correctly answer factual questions of a general nature. Teams from Orange Research, specialised in Data and Knowledge field, have created a system that gives a variety of answers, which will be presented at the Research Exhibition.

Une expérience enrichie avec le système Conversational search for general knowledge pour les utilisateurs qui posent des questions d’ordre général à leur majordome virtuel au sein de la maison.

Sales of domestic assistants are on the up, and many of us use them on a daily basis, mainly for three things: to find out the weather, to play a piece of music or to ask general knowledge questions. The team of researchers looked into the last usage type.

Improving the uses of Artificial Intelligence

“We quickly realised that there were some things voice assistants didn’t know when asked about general issues. They are generally only able to answer when they have the current answer to the question in their knowledge base and if possible in similar wording, interaction with them is rarely a real dialogue, and they only provide one answer, which is difficult to verify. Consequently we tried to build a system that improves these points,” explains Frédéric Herledan, Research Project Leader at Orange. “We worked on a wide variety of subjects, from a scientific and technical point of view, such as speech – and speaker – recognition, or dialogue with the machine.”

A system that gives smart, sensitive answers

Conversational search for general knowledge presents a system that answers factual general knowledge questions while helping users to better assess the relevance of the answers given. “To this end, it lists several alternative solutions according to Wikidata, and presents the most likely answer first, while giving access to the other possible answers. It can explain how it went about giving some of its answers, and provide excerpts from Wikipedia or Wikidata factsheets to expand the point of view on the issue,” Frédéric Herledan continues. “We also looked into the customisation of interaction and dialogue with the machine. Our system is able to identify and recognise speakers and to address them personally. What’s more, it knows how to link two questions asked one after the other, without all the keywords being repeated. For instance, you can ask “How old is Barack Obama?”, then “How old is his wife?” This of course implies that we are talking about Barack Obama’s wife. Dialogue with the system becomes fluid!”

A system that is 100% made in Orange

The Conversational search for general knowledge system uses a wide variety of application program interfaces (API) developed in-house by Orange. From the identification system and voice recognition to the fluid response provided by generation, the system moves from interface to interface to conduct a semantic analysis of the question, and to answer it using data from the knowledge base. Speech reproduction is then generated by a speech synthesis system, thanks to an Orange spin-off. This entire system could fit into a somewhat powerful home computer to meet the needs of users who are reluctant to transmit their data over the Internet, as is the case for existing virtual assistants.

Read also on Hello Future

In-Vehicle Commerce, a white paper co-authored by Orange and Worldline


Understand and Manage Your Digital Footprint at Home

Une développeuse avec son ordinateur devant une armoire informatique

AI Research for Discovering Faults in the Fiber Access Network


Can Chatbots Really Influence Us?


Behavioural objects in the home: laying the foundations


Personality derived from data can revolutionise smart services


Orange 2021 Research Exhibition: let’s prepare for the future!


Simulated data on household activity: we evaluate how realistic it really is!