On Wed, Feb 09, 2005 10:10 PM, "Michael Carl" <[EMAIL PROTECTED]> wrote:
As soon as discontinuous phrases (i.e. templates) will be used in SMT, indeed EBMT and SMT will be hard to distinguish
Indeed, this is already the case in the "Grammatical Channel" model of Wu & Wong from Coling/ACL-98, which used a SITG (stochastic inversion transduction grammar) to model the noisy channel so as to permit discontinuous phrases, by using the full range of typical syntactic constituents (and combinations thereof). (http://www.cs.ust.hk/~dekai/library/WU_Dekai/colacl19980520.ps)
Also perhaps useful: I discussed the convergence of the characteristics of SMT, EBMT, and Transfer MT in a panel talk at MT-Summit 2003. (http://www.cs.ust.hk/~dekai/library/WU_Dekai/WU_Dekai_MTSummit_20030726.pdf)
-Dekai
Dekai Wu / [EMAIL PROTECTED] / www.cs.ust.hk/~dekai HKUST, Human Language Technology Center Department of Computer Science University of Science & Technology, Clear Water Bay, Hong Kong tel +852 2358.7000 / dir +852 2358.6989 / fax +852 2358.1477
_______________________________________________ Mt-list mailing list
