Biomedical Engineering Reference
In-Depth Information
6.2.6
Motion Conclusion Summary
Physically actuated embodiment is certainly essential for robots to interface with the real world.
Additionally, many roboticists debate that interface with the complex, nonlinear real world is
important for the formation of intelligence (Brooks, 1991). Whether that is true or not, intelligence
and perceptive systems are of great importance for robots to be effective in the real world.
6.3
BEHAVIOR, EXPRESSIVITY
6.3.1
Intelligence and Perception
The emulation of human and animal central nervous systems (CNS) stands as the most challenging
domain of bio-inspired robotics. While neuroscience is deciphering the mysteries of mind at
unprecedented rates, thanks largely to novel imaging techniques such as fMRI, many components
of machine perception and intelligence are coming into functional maturity. Though not nearly
as capable as humans, many ''human-emulation'' technologies have sprouted substantially in the last
decade, showing remarkable surges in functionality including face tracking, feature tracking, visual
biometric identification, bipedal locomotion, and semantically rich NLP (Kurzweil, 1999; Menzel,
2001; Bar-Cohen and Breazeal, 2003). With these tools, we can sketch crude models of simulated
mind in technological media. The emphasis, however, is on the word ''crude'': it must be acknow-
ledged that most of the mysteries of the CNS are well beyond science at this time.
Accordingly, machine intelligence is decidedly below that of most animals and certainly
humans. But our machines must be judged on their own standards. After all, a machine can
understand speech better than a dog can, and what's more, the machine can speak back. Many of
the intelligent and perceptive systems available today have yet to be integrated into functional
whole. This section first considers intelligent systems as parts, and then discusses their integration
into a systematically emulated animal intelligence, with a focus on social intelligence.
6.3.1.1
Language, Ontologies, Top-Down
At the foundation of human-machine language interaction lie ASR, automated speech synthesis
(ASS), and various approaches to NLP. Although only capable of rudimentary language inter-
actions, machine language has recently shown a remarkable rate of progress, both in successful
academic research and in deployed speech solutions.
For many years, basic speech recognition and synthesis were major obstacles even to the most
elementary human-computer language interactions. However, progress in the late 1980s and 1990s
led to a large number of deployed speech applications, ranging from dictation software such as
IBM's Viavoice to natural language ticketing and customer service agents, such as those offered by
ATT. Companies now marketing commercial speech applications include SpeechWorks, Sensory,
Nuance, and Dragon Naturally Speaking. Another highly effective system is open-source to
researchers: Carnegie Mellon's Sphinx is highly functional, robust, user-independent ASR software
(Carnegie Mellon, website, 2002).
Several common features operate rather naturally; the ''barge-in'' capability allows users to
interrupt the system and still have speech recognized. ''Rejection and keyword spotting'' recognizes
a speaker's keywords without prompts. Using Bayesian analysis, ''N-best'' sorts through possibil-
ities of what a speaker might have said to locate a correct word, while the statistical language
modeling of ''N-gram'' creates a sizable vocabulary and natural language recognition.
Word recognition and synthesis is only the first step toward endowing machines with humanlike
language intelligence. Text-to-speech (TTS) software outputs increasingly natural-sounding speech,
with off-the-shelf solutions including Rhetorical, Elan, Nuance, and the open-source Festival.
Search WWH ::
Custom Search