Antw: Aberdeen MT system

[Nancy Frishberg]

Based only on these news reports, and experience over at least 20 years with similar "successes" I predict Franz is correct in his assessment.

I have seen programs that can recognize single signs (from a vocabulary of >25 items, albeit with minimal pairs among them).
I have seen demonstrations of recognition of handshapes (like fingerspelling).

But I have not seen any lab that is able to recognize fluent sign language at speed and without regard to topic, even to be able to recreate it on screen (whether by animation or something similar) without the complication of translation. When I talk to computer scientists about this process, I describe it as the confluence of machine vision, machine translation, "speech" to text, at least, and if it were going to replace a live (local or remote) interpreter, it would also involve text to speech (which is a more well-understood and nearly solved problem). And of course, do we have a sufficiently robust grammar of BSL (or ASL) to tell a machine how to parse the syntax? (e.g., where do gaze and body position figure into this scheme?)

But I'd be delighted to be surprised, and will poke among my other resources to see if I can find out more about the Aberdeen project.

 - Nancy

On Mar 13, 2012, at 12:39 AM, Franz.Dotter at UNI-KLU.AC.AT wrote:

> Dear colleagues,
>  
> After having read the news text and other sources on PSLT, I suppose that this system tries to identify single handshapes in isolated production as "signs" (there is nowhere a notice concerning movement, but it maybe included?). As we had some years ago a similar sensational message which turned out that the software could recognise the finger alphabet letters, my associative interpretation is: These colleagues have adopted the idea of recognising finger alpahbet letters and expanded it to a (how much limited?) system where any user can "teach" new signs to the system.
> My inpression is that the Scottish colleagues don't cope with syntax, neither with the coarticulative change of signs but believe in "sign language is a langaue where the single signs are produced separatedly (i.e. with some pause/release between) one after the other without such "disturbances" like syntax.
>  
> Moreover, the text of the announcement contains some very doubtful passages like:
> "BSL is however, a general-purpose language and therefore poses limitations for users, making it impossible for them to easily express certain concepts and terms that are very specific or used only within particular areas of society – for example education and the workplace.
> To overcome this, PSLT enables users to personalise sign language to their own individual needs.
> 
> Dr Compatangelo continued: “One of the most innovative and exciting aspects of the technology, is that it allows sign language users to actually develop their own signs for concepts and terms they need to have in their vocabulary, but they may not have been able to express easily when using BSL."
> 
> The colleagues should be asked  how many lexical items a sign language has in their opinion. If I'm right with my suspicion, they may be surprised by the fact that SLs may contain several thousands of signs.
> 
> To recognise somehow defined isolated hand gestures or similar, may have a positive impact on the inclusion of SL users in the heraing society, but the developers should be honest about the limits of their system.
> 
>  
> Best Regards
> 
> Franz Dotter
> 
>  
> 
> 
> >>> "Dan Parvaz" <dparvaz at GMAIL.COM> 3/13/2012 12:24 >>>
> Has anyone seen the system mentioned in the URL below? And has there been 
> any systematic testing of this Machine Translation system which, according to 
> _The Scotsman_, would "dramatically change the lives of more than 50,000 users 
> of British Sign language whose communication would be 'immediately translated 
> into text which can be read by the person they are conversing with'"?
> 
> (Be that as it may, I haven't seen Spanish MT dramatically change the far greater 
> number of Spanish-speakers in the US, but that may just be me being grumpy)
> 
> Just curious.
> 
> Cheers,
> 
> -Dan
> 
> PS: Here's the URL: http://www.telegraph.co.uk/science/science-
> news/9134827/Sign-language-program-converts-hand-movements-into-text.html
> 

Nancy Frishberg +1 650 804 5800 mobile
nancyf at fishbird.com
Twitter: @nancyf
Blog: www.fishbird.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/slling-l/attachments/20120313/90b8787f/attachment.htm>