Manifesto
In machines that converse with people there is a very well-drawn map and a whole territory left blank.
The map covers the standard user: the one who pronounces as expected, asks for what was foreseen and answers when it is their turn.
That is where almost everyone works, and there the systems work.
The blank territory is everything else.
The strong accent the recogniser cannot decipher.
Speech affected by a disorder.
The voice of an eighty-year-old.
The language with few resources.
And also the request that arrives in a muddle, the sentence corrected halfway through, the silence that lasts longer than the system can bear, the conversation that needed repairing and was restarted from scratch.
It is not blank because it is impossible. It is blank because it is expensive, slow and makes fewer headlines.
We have worked there since 2005.
It is a decision, not a limitation.
We are a small group because we are focused on one thing only: machines that converse with real people. That takes understanding what was said, what was meant and when to answer — all three, or there is no conversation.
We do not do artificial intelligence in general. We do the hard part of one specific thing, and we have been doing it for twenty years.
We go with rigour.
European projects, peer review, international collaboration.
We come back with something people use.
In a consulting room, at a service counter, in a public service.
The map is already drawn.
What is interesting lies outside.