Dear support,

follow-up on my previously submitted error description: after an hour
of searching through Moses source code I found out that the problem
was that i confused the factor (0) and LM order (3) while specifying
the LM for training (thanks to the power of copy-paste it was the same
on both computers, all configurations and corpora). Thus in
Hypothesis.cpp on the 277th row in Hypothesis::CalcLMScore(LMList) it
obtained 0 as the LM order, and since the order it's an unsigned int,
on the 308th row (for loop from 0 to nGramOrder-1) it looped to
2^32-1.

Concluding: if LM order 0 should be allowed (constant weight to all
words) then it doesn't work. If it shouldn't be allowed, then Moses
should display some error message in the beginning (erroneous LM order
or s.t).

I would fix it myself but that would mean another hour searching for
the appropriate place to put the LM order checking into, I'm sure the
original developers can find the spot faster.

Best regards,

Mark Fishel
Dept. of Computer Science
University of Tartu

On Fri, Oct 3, 2008 at 12:19 PM, Mark Fishel <[EMAIL PROTECTED]> wrote:
> Dear support,
>
> after updating moses to the latest version via SVN, moses dies with a
> segmentation fault while translating the first sentence -- i.e.
> regardless of used corpus or configuration, as it seems. The last
> lines it utters before dying are
>
> added hyp to stack, best on stack, now size 1
> processing hypothesis from next stack
> Segmentation fault
>
> Full log of "fully verbal" (-v 100) moses is attached. Any help appreciated.
>
> Best regards,
>
> Mark Fishel
> Dept. of Computer Science
> University of Tartu
>
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to