[Moses-support] zlib bug causing Moses training to fail
Hi I've discovered that an apparent bug in zlib (specifically version 1.2.3.4 installed in Ubuntu 12.04) can cause the Moses training pipeline to fail. Upgrading to the latest zlib (1.2.8) fixes the problem. The problem appears when the extract files are being read in order to create the reordering tables or translation tables. For the reordering table, an error message like this terminate called after throwing an instance of 'util::GZException' what(): util/read_compressed.cc:163 in virtual std::size_t util::{anonymous}::GZip::Read(void*, std::size_t, util::ReadCompressed) threw GZException'. zlib encountered invalid distances set code -3 is printed, whilst for the translation table, the Moses scorer swallows the zlib error but eventually fails in 'consolidate' like this: ERROR: source phrase does not match in line 289: 'clean .' != 'definitely recommend to have' So far I haven't found any reports of this zlib bug, and it seems to be quite intermittent. cheers - Barry -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
Re: [Moses-support] Prereordering was: Re: Google Translate - technology paradigms behind it
Not that I know of. If you create 1, please check it in. On 09/01/2014 15:49, Per Tunedal wrote: Hi, are there any tools to do prereordering with Moses? I suppose it has to be done both before training and translation. And that it presupposes tagging. How much does prereordering improve translation? Yours, Per Tunedal On Wed, Jan 8, 2014, at 22:42, Philipp Koehn wrote: Hi, Google lists its research papers here: http://research.google.com/pubs/MachineTranslation.html --- snip--- maybe hierarchical, with pre-reordering. ---snip --- -phi On Wed, Jan 8, 2014 at 8:47 PM, Ivan Dun?er ivandun...@gmail.com mailto:ivandun...@gmail.com wrote: Hi everybody, is there any reading material (e.g. scientific articles etc.) on what MT approaches Google Translate uses? Sorry for off-topic. Regards, Ivan ___ Moses-support mailing list Moses-support@mit.edu mailto:Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support _ Moses-support mailing list Moses-support@mit.edu mailto:Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
Re: [Moses-support] Moses-support Digest, Vol 87, Issue 20
I know Jakob Elming at UNICPH did prereordering before training MT models with Moses. He might be able to help. -Andreas 2014/1/10 moses-support-requ...@mit.edu 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-requ...@mit.edu You can reach the person managing the list at moses-support-ow...@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. static build of mosesdecoder (Tak Kuya) 2. zlib bug causing Moses training to fail (Barry Haddow) 3. Re: Prereordering was: Re: Google Translate - technology paradigms behind it (Hieu Hoang) -- Message: 1 Date: Fri, 10 Jan 2014 18:56:33 +0900 From: Tak Kuya kuya...@gmail.com Subject: [Moses-support] static build of mosesdecoder To: moses-support@mit.edu Message-ID: CANaijngK2Q2QpaQ4CPP7z5= wzrza48gmphukgcvppy2sr6g...@mail.gmail.com Content-Type: text/plain; charset=iso-8859-1 Hi. How do you build mosesdecoder using static libraries like http://www.statmt.org/moses/RELEASE-1.0/binaries/ ? To make things simple, I use the master branch instead of Rel.1.0 and edit mosesdecoder/Jamroot a little and build the decoder only. I can build the normal shared version of moses without any problems and usually use it (on CentOS 6.5 with boost 1.41). But I failed to build moses with --static option. So I installed boost 1.41 to my home directory. But I still couldn't build moses. It said cannot find -lz and -lrt so I installed zlib-static and glibc-static with yum. And... It failed again. Here are build.log and Jamroot I used. Could anyone teach me how to solve the problem? Thank you. Best regards, Tak Kuya -- next part -- A non-text attachment was scrubbed... Name: build.log Type: application/octet-stream Size: 2184 bytes Desc: not available Url : http://mailman.mit.edu/mailman/private/moses-support/attachments/20140110/03806883/attachment-0002.obj -- next part -- A non-text attachment was scrubbed... Name: Jamroot Type: application/octet-stream Size: 6038 bytes Desc: not available Url : http://mailman.mit.edu/mailman/private/moses-support/attachments/20140110/03806883/attachment-0003.obj -- Message: 2 Date: Fri, 10 Jan 2014 14:13:57 + From: Barry Haddow bhad...@staffmail.ed.ac.uk Subject: [Moses-support] zlib bug causing Moses training to fail To: moses-support@mit.edu Message-ID: 52d00025.9090...@staffmail.ed.ac.uk Content-Type: text/plain; charset=ISO-8859-1; format=flowed Hi I've discovered that an apparent bug in zlib (specifically version 1.2.3.4 installed in Ubuntu 12.04) can cause the Moses training pipeline to fail. Upgrading to the latest zlib (1.2.8) fixes the problem. The problem appears when the extract files are being read in order to create the reordering tables or translation tables. For the reordering table, an error message like this terminate called after throwing an instance of 'util::GZException' what(): util/read_compressed.cc:163 in virtual std::size_t util::{anonymous}::GZip::Read(void*, std::size_t, util::ReadCompressed) threw GZException'. zlib encountered invalid distances set code -3 is printed, whilst for the translation table, the Moses scorer swallows the zlib error but eventually fails in 'consolidate' like this: ERROR: source phrase does not match in line 289: 'clean .' != 'definitely recommend to have' So far I haven't found any reports of this zlib bug, and it seems to be quite intermittent. cheers - Barry -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. -- Message: 3 Date: Fri, 10 Jan 2014 15:05:45 + From: Hieu Hoang hieuho...@gmail.com Subject: Re: [Moses-support] Prereordering was: Re: Google Translate - technology paradigms behind it To: moses-support@mit.edu Message-ID: 52d00c49.7010...@gmail.com Content-Type: text/plain; charset=iso-8859-1 Not that I know of. If you create 1, please check it in. On 09/01/2014 15:49, Per Tunedal wrote: Hi, are there any tools to do prereordering with Moses? I suppose it has to be done both before training and translation. And that it presupposes tagging. How much does prereordering improve translation? Yours, Per Tunedal On Wed, Jan 8, 2014, at 22:42, Philipp Koehn wrote: Hi, Google lists its research papers here: http://research.google.com/pubs/MachineTranslation.html --- snip--- maybe hierarchical, with pre-reordering. ---snip --- -phi On Wed, Jan 8
Re: [Moses-support] static build of mosesdecoder
double check that libz.a and librt.a exist in /usr/lib or /usr/lib64 I statically compiled on Suse Linux and it was ok. My bjam command was: ./bjam --with-boost=/home/s0565741/workspace/boost/boost_1_54_0/ --with-irstlm=/home/s0565741/workspace/irstlm/ --with-cmph=/home/s0565741/workspace/cmph-2.0 --with-dalm=/home/s0565741/workspace/github/DALM -j16 -static -a On 10/01/2014 09:56, Tak Kuya wrote: Hi. How do you build mosesdecoder using static libraries like http://www.statmt.org/moses/RELEASE-1.0/binaries/ ? To make things simple, I use the master branch instead of Rel.1.0 and edit mosesdecoder/Jamroot a little and build the decoder only. I can build the normal shared version of moses without any problems and usually use it (on CentOS 6.5 with boost 1.41). But I failed to build moses with --static option. So I installed boost 1.41 to my home directory. But I still couldn't build moses. It said cannot find -lz and -lrt so I installed zlib-static and glibc-static with yum. And... It failed again. Here are build.log and Jamroot I used. Could anyone teach me how to solve the problem? Thank you. Best regards, Tak Kuya ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
[Moses-support] Moses kbmira/pro with non-Moses decoder
Hi, What's the best way to use Moses optimization without using Moses the decoder? I've got my own system combination decoder (MEMT). Currently, the scripts use Z-MERT but 1) MERT supports a limited number of features and 2) Users keep finding bugs in Z-MERT. I can run pro or kbmira directly but is there documentation of their file formats? Another option is to use mert-moses.pl and try to look like Moses. How much does my decoder have to look like Moses for that to work? Keep in mind the input isn't a flat text file anymore. Kenneth ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
[Moses-support] does mosesserver use -threads argument?
Hi all, Does anyone know whether mosesserver use -threads argument to control the number of threads? As we know that mosesserver uses XMLRPC to set up a http server, and I found that XMLRPC only has two modes about threads: serial (using runOnce method) or unlimited threads (using run method) from: http://xmlrpc-c.sourceforge.net/doc/libxmlrpc_server_abyss++.html I had a look at moseserver.cpp, and I think mosesserver does not use the -threads to control the number of threads. Is anyone familiar with this? Thank you in advance! Best wishes! -- Wang Pidong Department of Computer Science School of Computing National University of Singapore ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
Re: [Moses-support] does mosesserver use -threads argument?
Hi Wang Yes, you're right. mosesserver uses the threading model from xmlrpc, which is either a thread per connection, or single threaded. cheers - Barry Quoting Wang Pidong wan...@comp.nus.edu.sg on Fri, 10 Jan 2014 12:40:36 -0800: Hi all, Does anyone know whether mosesserver use -threads argument to control the number of threads? As we know that mosesserver uses XMLRPC to set up a http server, and I found that XMLRPC only has two modes about threads: serial (using runOnce method) or unlimited threads (using run method) from: http://xmlrpc-c.sourceforge.net/doc/libxmlrpc_server_abyss++.html I had a look at moseserver.cpp, and I think mosesserver does not use the -threads to control the number of threads. Is anyone familiar with this? Thank you in advance! Best wishes! -- Wang Pidong Department of Computer Science School of Computing National University of Singapore -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
Re: [Moses-support] Moses kbmira/pro with non-Moses decoder
Hi Kenneth If your decoder can output the n-best list in the same format as Moses, then you can run the Moses extractor (to get bleu sufficient stats and features from the n-best), then the optimiser. For PRO this means running the Moses pro binary, followed by MegaM. For k-best mira, I think you just have to run the kbmira binary. You should be able to see what arguments to use from looking at the output of a mert-moses.perl run, as it reports the commands as it runs them. Each of the optimisers outputs the new weight set, which you then have to parse and send back to your decoder. I don't think these formats are documented, but they should be clear enough. If you want to make your own decoder look like Moses, then the outputting the n-best list is probably straightforward - you have to do that anyway. The hassle is that mert-moses.pl will expect your decoder to understand all the Moses command line options for specifying new weights, cheers - Barry Quoting Kenneth Heafield mo...@kheafield.com on Fri, 10 Jan 2014 12:39:09 -0800: Hi, What's the best way to use Moses optimization without using Moses the decoder? I've got my own system combination decoder (MEMT). Currently, the scripts use Z-MERT but 1) MERT supports a limited number of features and 2) Users keep finding bugs in Z-MERT. I can run pro or kbmira directly but is there documentation of their file formats? Another option is to use mert-moses.pl and try to look like Moses. How much does my decoder have to look like Moses for that to work? Keep in mind the input isn't a flat text file anymore. Kenneth ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. ___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
[Moses-support] How to Improve the translation
Hi All, I've developed Urdu to English SMT using MOSES, and it is currently giving me BLEU score of 8. Now I wanted to improve its translation so that it gives me higher BLEU score. Regards Asad A.Malik___ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support