naturerecordists
[Top] [All Lists]

Predicting finch calls [slightly OT]

Subject: Predicting finch calls [slightly OT]
From: "gweddig" gweddig
Date: Wed Nov 17, 2010 5:52 am ((PST))
>From Wired News

http://www.wired.com/wiredscience/2010/11/song-bird-language/

Text follows:

The notoriously complex song of the finch has finally succumbed to statisti=
cs. Physicists have developed a model that can map out and predict the note=
s birds sing in sequence.

The new model is roughly eight times more accurate than previous attempts t=
o unravel complex bird songs. If the scientists can use the same technique =
to map and predict chatter in other social animals, they could find importa=
nt clues about the neural origins of complex language, including that of hu=
mans.

"We want to gain an understanding of this simplest case, then work our way =
up in complexity," said physicist Dezhe Jin of Penn State University who le=
d the research, posted Nov. 12 on arXiv.org.

Birdsong originates at the top of a bird's brain in an area called the HVC,=
 or higher vocal center, which is made up of about 40,000 neurons. Networks=
 of thousands of individual neurons there are thought to generate syllables=
, and these neural networks link up to other areas of the brain to actually=
 vocalize the sounds.

Mapping the sounds and their sequence in a song may help resolve such langu=
age-centric brain pathways.

"We think it's like a domino effect, where one syllable cascades into the n=
ext to create complex songs," Jin said. "But before neural coordinates can =
be verified, we need to have robust statistical maps."

Jin stuck a Bengalese finch in a soundproof room for six days with a microp=
hone. The bird tweeted more than 25,000 times, sounds that Jin and his team=
 divvied up into 25 groups based on statistical similarity. In total, the f=
inch sang seven distinct song syllables (sounds made very quickly one after=
 the other) and 14 other types of notes.

Unlike previous models, which skyrocket in error when trying to predict mor=
e than one note in sequence, the new model factors in the order of previous=
 notes. It also takes into account the fact that different neural networks =
may produce the same syllable which, Jin says, provides a subtle but crucia=
l detail in correctly mapping and predicting a song's syllables.

No model will ever be able to predict a bird's song with 100 percent accura=
cy because they improvise as they go, like jazz musicians, Jin said. But th=
ey may be able to get close enough to begin to understand what's happening =
in the bird's brain.

"This is really the beginning of finding how song and language structure or=
iginates," Jin says. "We want to further study other species and apply that=
 knowledge to humans."









<Prev in Thread] Current Thread [Next in Thread>
  • Predicting finch calls [slightly OT], gweddig <=
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU