Spreading Activation, Lexical Priming and the Semantic Web by Michael Pace-Sigge

Spreading Activation, Lexical Priming and the Semantic Web by Michael Pace-Sigge

Author:Michael Pace-Sigge
Language: eng
Format: epub, pdf
Publisher: Springer International Publishing, Cham


It is this point that shows how the discussion can close the circle fully: because the training data are so highly (domain) specific, it gives the neural network machine the chance to learn the properties of the language as received per input. As a result (an unintentional result in this case), the machine opts to apply the fundamental ground rules thus learned. Et voilà: a machine is born that sounds like a natural language producer, rather than a dumb parrot. In fact, as this book is in its production stage, Google, during its Developer’s Conference, unveiled Google Duplex , the latest development of their AI-driven voice assistant products. Duplex appears more natural to conversation partners as the typical disfluencies of natural speakers are integrated. The result is “...a new technology for conducting natural conversations to carry out “real world” tasks over the phone. The technology is directed towards completing specific tasks, such as scheduling certain types of appointments.” The authors describe the technology as follows: “at the core of Duplex is a recurrent neural network (RNN) designed to cope with these challenges, built using TensorFlow Extended (TFX). To obtain its high precision, we trained Duplex’s RNN on a corpus of anonymized phone conversation data”. (Leviathan and Matias 2018). The demonstration appears, indeed, disconcertingly human-like, understanding typical parts of a service exchange (“hold on one second”) while integrating typical discourse particles and back channelling markers (“Hi, um, I’d like” … “Mm hmm”) in their conversation. 30



Download



Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.