Monday, November 16, 2009

Monday Science Links

Here's this week's science:
  • Martin at Aarvachaeology offers us a silver figurine from storied Lejre in Denmark: She's wearing a floor-length dress. And a shawl. And four finely sculpted bead strings. This is a standard depiction of an aristocratic lady of the later 1st Millennium. The Lejre figurine is a direct counterpart to the Aska pendant (below), which is universally understood as the effigy of a goddess. The high seat is Odin's, allright. But the occupant is most likely Frigga or Freya. Or maybe, just maybe, Thor in drag during the hammer reclamation mission. That is so cool! This find will mess with everybody's mind!

  • Kim at All Of My Faults Are Stress Related discusses earthquake triggers vs. causes: There have been a number of studies in the past decade or so that suggest that earthquakes can be triggered by little things, such as the passage of seismic waves. The studies are fascinating, in part because the triggers seem so small in comparison to any other force (like the weight of the rock). How could such a little thing unleash an earthquake?

  • David at Irregular Web Comic muses on computer voice recognition: I know it's a hard problem to tackle from a computer science point of view. But I can't help feeling that we are puny ants on the face of an edifice of such size and elegance that we can't discern the patterns for which we seek. That computer science is tackling the problem of voice recognition in completely and utterly the wrong way.

  • Ed at Not Exactly Rocket Science tells the story of FOXP2, the 'language gene': Several years on, and it is clear that talk of a "language gene" was premature and simplistic. Nevertheless, FOXP2 tells an intriguing story. "When we were first looking for the gene, people were saying that it would be specific to humans since it was involved in language," recalls Simon Fisher at the University of Oxford, who was part of the team that identified FOXP2 in the KE family. In fact, the gene evolved before the dinosaurs and is still found in many animals today: species from birds to bats to bees have their own versions, many of which are remarkably similar to ours. "It gives us a really important lesson," says Fisher. "Speech and language didn't just pop up out of nowhere. They're built on very highly conserved and evolutionarily ancient pathways."

  • And at The White Coat Underground, PalMD looks at the resurgence of mumps: The resurgence of vaccine-preventable diseases is a fascinating, if unwanted, phenomenon. Pertussis, measles, and now mumps are cropping up after long periods of quiescence. Mumps has been generally very well-controlled since the adoption of wide-spread vaccination, with no nation-wide outbreaks, but there have been a number of regional outbreaks, most notably in 2006 and now again in 2009. ... The vaccination rate among those infected was significantly lower than the average for New York state, but still somewhat high (around 72% for those in whom vaccination status was known). This is below herd immunity rates, but obviously raises questions.
Enjoy!

Labels: , ,

3 Comments:

At 7:12 AM, November 16, 2009 Blogger Barry Leiba had this to say...

[This has gotten long; sorry. Maybe I should put it up as a main entry on my blog.]

The dog vs computer joke is cute, but it isn't science. The fact is that computers can reliably recognize voice commands at the same level as a dog can: they can both recognize what they're trained to recognize.

For things they aren't trained to recognize, they can also do very well if they're given enough samples of the speaker's voice to customize their pattern analysis. In fact, a speech-reco program that's got a bunch of speakers in its database can even figure out which one is speaking, and then do a pretty accurate job of getting the words right.

Some 20 years ago, I was demonstrating IBM's ViaVoice system to someone, just off the cuff — it was on the spur of the moment, and I wasn't prepared. I picked up a nearby novel and read the first paragraph. It got it with 100% accuracy. 100%. I was quite surprised.

Of course, I speak clearly, and the program had been taught my voice (and no others). Still, the point is that the techniques we're using are effective at a lot of things. But, no, they can't come close to what a five-year-old can do with her ears and brain.

David says that "it wouldn't surprise me in the least if some young gun came along next year and did something completely out of left field that nobody in the research landscape had even considered before, and it turns out to vastly simplify the problem to something that is actually tractable to our computers."

Well, yes... that's the sort of thing that happens in science. All the time. Possibly moreso in computer science (which is newer than, say, biology or physics), but in every field of science, generally. And we love it when it happens. We're treating speech recognition as a pattern-matching problem. We don't know what the brain does, of course, and someone certainly may flash an "Aha!" that gives us a different and more effective way to approach it.

But right now, the real problems come when we try to have arbitrary voices say relatively arbitrary things. Right, when you want a pizza delivered to "Phillip Street", and you pronounce it more like "Phelp", and it turns out that it really should have been "Avenue", but, well, a human would have figured it out... should anyone really be surprised when the computer can't get it?

How often do you have to spell your street name to a human, even one who's older than five?

 
At 8:30 AM, November 16, 2009 Blogger The Ridger, FCD had this to say...

I might tend more to your point of view if the "tell me your address" phone thingies weren't so damned bad at recognizing them when they hear them. Heck, one last week didn't understand "Maryland" until I said " MARE-EE-LAND" to it, slowly, and there are only a relatively few states it could have been!

But you're right, of course. David's making a joke.

 
At 7:33 PM, November 16, 2009 Blogger Barry Leiba had this to say...

Oy. The "Maryland" one should have been easy: a domain of only 50 (or 51, with DC, or 55, with the territories as well) shouldn't be hard for it to deal with. There are good programs, and crappy programs. And we also don't know what the audio sounded like on the other end.

Anyway, sorry: I was only picking at it because you filed it under "science".

 

Post a Comment

Subscribe to Post Comments [Atom]

     <-- Older Post                     ^ Home                    Newer Post -->