[Mt-list] Re: Mt-list Digest, Vol 1, Issue 3

[Andy Way]

After reading the introduction, almost all examples of what they call 
Phrasal SMT seems (to me) examples of EBMT systems.

I think Simon's reply answers your question. Nonetheless (and picking up 
a theme
I started on the list last summer), I find it pretty much incredible 
that you won't find
the words 'example-based' in _any_ paper on Phrase-based EBMT. In a 
recent paper
comparing our Marker-Based EBMT system to SMT (which'll appear in the 
special
issue of 'Language Engineering' that Michel & Rada are editing), I noted 
that no
phrase-based SMT papers acknowledge the (obvious, to me) debt they owe 
to EBMT:
after all, EBMT has been developing sub-sentential syntactic and lexical 
chunks for
years before the SMT thought it'd be a good thing to try. I don't know 
why this is.

Very occasionally, you'll hear commentators (such as Ed Hovy at the MT 
Summit IX
panel session) say that what (say) Franz Och is doing is EBMT-like, but 
nowhere will
you see this written down. I'd love to be corrected here, but I've read 
most of the SMT
papers in recent years and haven't seen it anywhere.

As an aside, I think the phrasal chunks that SMT produces are different 
from EBMT
chunks in that they're pretty much just n-grams, whereas EBMT systems 
derive their
sub-sentential alignments on the basis of (at least some) syntactic 
reasoning.
Andy.

_______________________________________________
Mt-list mailing list