SIRI
RISING
of speech in a sentence. But Siri
abandoned that method in favor of a breakthrough approach
devised by Cheyer and his colleagues. Instead of modeling linguistic concepts, their system
could model real-world objects.
Told, “I want to see a thriller,”
Siri would immediately identify
“thriller” as a film genre — and
summon up movies — rather than
analyze how the subject connected to an object and a verb.
Siri was able to map the contents of a question onto a domain
of potential actions, then pick the
action that seemed most probable,
based on its understanding of the
relationships between real-world
concepts. (For example, Siri knew
a given restaurant should have a
rating, an address, a type of cuisine and a price range associated
with it.) Siri could also apply details about the time of day and a
user’s preferences and location to
inform its response, or to ask for
more information.
Picture Siri as a concierge in
a noisy lobby. A request for the
“closest coffee shop” might sound
like “closest call Felicia” over the
din. But knowing that “closest” is
more likely to characterize a place
than a person, and that a guest is
HUFFINGTON
01.27.13
more likely ask a concierge for dining tips, a human hotelier would
infer the asker was probably hankering for a cappuccino. Same
with Siri, which was tuned to listen for ѡ