For French-English translation, hierarchical models will probably not
do any better than phrase-based models.  The rule-of-thumb seems to be
that language which large-scale reordering fare better with
hierarchical models, but language pairs with only relatively local
reordering are better with phrase-based systems.

>
> Results (bleu scores) for my hierarchical phrase model are slightly
> worse than using the good old phrase based decoder. Does anybody have
> experiences if tagging the (English) source with an annotating parser
> (bitpar and collins are mentioned in the documentation) improves things?
>
> best regards
> Christof
>
>
>
>
>
>>
>> On 01/02/2010 19:13, Christof Pintaske wrote:
>>> Hi,
>>>
>>> while running mert-moses-new.pl to tune the chart-decoder I get this output:
>>>
>>> [...]
>>> The decoder returns the scores in this order: lm tm tm tm tm tm tm w
>>> [...]
>>> Executing: /export/home/moses/libexec/moses-chart/bin/extractor
>>> --scfile run1.scores.dat [...]
>>> The decoder produced also some 'tm' scores, but we do not know the
>>> ranges for them, no way to optimize them
>>>
>>> it seems it dies on these 'tm' scores. Is there a way to prevent the
>>> decoder from producing these scores?
>>>
>>>
>>> regards
>>> Christof
>>>
>>>
>>>
>>> Christof Pintaske wrote:
>>>
>>>> Hi,
>>>>
>>>> I'm running
>>>>
>>>> mert-moses-new.pl
>>>>    ~/en-fr_chart/tuning/token_lowercase.en
>>>>    ~/en-fr_chart/tuning/token_lowercase.fr
>>>>    /export/home/moses/libexec/moses-chart/bin/moses_chart
>>>>    ~/en-fr_chart/training/pm/model/moses.ini
>>>>    --mertdir=/export/home/moses/libexec/moses-chart/bin
>>>>    --working-dir=~/en-fr_chart/tuning
>>>>    --decoder-flags=-v 0 --no-filter-phrase-table
>>>>
>>>> for tuning. It performs exactly one iteration and writes a
>>>> run1.moses.ini in the tuning directory. That moses.ini is almost
>>>> identical to the one that I got after training (see below). Am I missing
>>>> the obvious?
>>>>
>>>> any hints are welcome
>>>>
>>>> Christof
>>>>
>>> _______________________________________________
>>> Moses-support mailing list
>>> [email protected]
>>> http://mailman.mit.edu/mailman/listinfo/moses-support
>>>
>>>
>>>
>> _______________________________________________
>> Moses-support mailing list
>> [email protected]
>> http://mailman.mit.edu/mailman/listinfo/moses-support
>
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
>

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to