Send Moses-support mailing list submissions to
moses-support@mit.edu
To subscribe or unsubscribe via the World Wide Web, visit
http://mailman.mit.edu/mailman/listinfo/moses-support
or, via email, send a message with subject or body 'help' to
moses-support-request@mit.edu
You can reach the person managing the list at
moses-support-owner@mit.edu
When replying, please edit your Subject line so it is more specific
than "Re: Contents of Moses-support digest..."
Today's Topics:
1. Re: Hyerarchial output search graph (Philipp Koehn)
2. Re: Per-sentence models (Prashant Mathur)
----------------------------------------------------------------------
Message: 1
Date: Thu, 10 Nov 2016 13:26:21 -0500
From: Philipp Koehn <phi@jhu.edu>
Subject: Re: [Moses-support] Hyerarchial output search graph
To: Guillem Torres Badia <guitorba@inf.upv.es>
Cc: "moses-support@mit.edu" <moses-support@mit.edu>
Message-ID:
<CAAFADDBYRT-2qDphqKeAoE+nJNnYPYcXS0bP2o0wY22Er2CUpQ@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"
Hi,
I am not sure what you are asking here.
The output-search-graph features allows you to get the search graph
for each sentence. This additional output does not require any other
setting in the decoder.
-phi
On Tue, Nov 8, 2016 at 5:21 PM, Guillem Torres Badia <guitorba@inf.upv.es>
wrote:
> Hello,
>
> I am training an output-search-graph/hypergraph with MOSES chart. I would
> like to know which are the training parameters I should consider and their
> values that usually give best results in BLEU.
>
> I attach you the script I'm using currently, just in case you need to
> check some details.
>
> Guillem
>
> _______________________________________________
> Moses-support mailing list
> Moses-support@mit.edu
> http://mailman.mit.edu/mailman/listinfo/moses-support
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20161110/93704c4d/attachment-0001.html
------------------------------
Message: 2
Date: Fri, 11 Nov 2016 14:18:14 +0100
From: Prashant Mathur <prashant@fbk.eu>
Subject: Re: [Moses-support] Per-sentence models
To: Lane Schwartz <dowobeha@gmail.com>
Cc: "moses-support@mit.edu" <moses-support@mit.edu>
Message-ID: <6E24AACB-AA5D-4403-AF88-884BAF9486A6@fbk.eu>
Content-Type: text/plain; charset="utf-8"
Hi Lane and all,
For our use case at eBay, we added a PhraseDictionaryCache for each sentence. It allows you to add phrase pairs with defined number of scores for each sentence.
Input with 3 features will look like this:
<seg id=?1?> <dlt type="cbtm" cbtm="das ||| the ||| 1.0 0.5 0.6 |||| ist ||| is ||| 1 1 1 |||| ein kleines haus ||| a cute place ||| 0.5 0.9 0.7"> das ist ein kleines haus </seg>
With mosesserver you need the segment IDs otherwise not; its because server puts all the segment ID to 0.
Its thread safe at segment level. But, if you input two segments with same ID there might be an issue.
In moses config, you just have to add this line:
'''
[feature]
PhraseDictionaryCache name=CBPT0 table-limit=20 num-features=3 input-factor=0 output-factor=0
[weight]
CBPT0= 0.3 0.3 0.3
'''
It also work with multiple factors (haven?t tested this for all possible corner cases), see the regression test here:
https://github.com/ebay-hlt/moses-regression-tests/tree/master/tests/phrase.cache-multi-factor <https://github.com/ebay-hlt/moses-regression-tests/tree/master/tests/phrase.cache-multi-factor>
Code is available here: https://github.com/ebay-hlt/mosesdecoder/ <https://github.com/ebay-hlt/mosesdecoder/>
Let me know if you have any issues.
Cheers,
?Prashant
On Mon, Nov 7, 2016 at 6:10 PM, Lane Schwartz <dowobeha@gmail.com <mailto:dowobeha@gmail.com>> wrote:
Hi,
I'm interested in extending Moses to enable per-sentence LMs and TMs. Given the current moses architecture (esp. StaticData), how feasible would it be to add new LMs and TMs that only exist for one sentence translation?
Thanks,
Lane
_______________________________________________
Moses-support mailing list
Moses-support@mit.edu <mailto:Moses-support@mit.edu>
http://mailman.mit.edu/mailman/listinfo/moses-support <http://mailman.mit.edu/mailman/listinfo/moses-support>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/mailman/private/moses-support/attachments/20161111/f8e1f7fa/attachment-0001.html
------------------------------
_______________________________________________
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support
End of Moses-support Digest, Vol 121, Issue 27
**********************************************
Subscribe to:
Post Comments (Atom)
0 Response to "Moses-support Digest, Vol 121, Issue 27"
Post a Comment