Hi Hieu,
Thanks. I haven't tried to use KenLM and IRSTLM. I will consider using KenLM or IRSTLM. Which language model out of these two LMs is better from your opinion? Regards, Lina _____ From: hieuho...@gmail.com [mailto:hieuho...@gmail.com] On Behalf Of Hieu Hoang Sent: 28 May 2013 17:15 To: Lina Jin Cc: Moses Subject: Re: [Moses-support] UnkownPtImplementation with AL suffix arrays i re-integrated SRILM into Moses last week, so git pull to get it https://github.com/moses-smt/mosesdecoder/commit/41da56364565e0aa9d40cce018e 5ef82f9766430 can i ask why you don't use KenLM or IRSTLM? I don't see any advantages using SRILM? On 28 May 2013 17:09, Lina Jin <l...@alphacrc.com> wrote: Hi Hieu, I notice that you commented out following line in moses.filtered.ini.1. #KENLM lazyken=0 name=LM0 factor=0 path=/home/waziz/experiments/wmt13/lm/interpolated-binlm.1 order=3 #LM0= 0.5 I tried to run EMS with target syntax model by using latest Mosesdecoder. Srilm language model is used. But it failed at the tuning stage because it can not recognise the feature "SRILM" in the section "# feature functions" of moses.filtered.ini.1 I tried to figure out what is the cause for it. Maybe you know the answer. Thanks in advance. Lina _____ From: moses-support-boun...@mit.edu [mailto:moses-support-boun...@mit.edu] On Behalf Of Hieu Hoang Sent: 28 May 2013 16:21 To: Wilker Aziz Cc: Moses Subject: Re: [Moses-support] UnkownPtImplementation with AL suffix arrays ok. done. I few provisos when you're using it: 1. Doesn't support multi-threading. Otherwise throws an exception 2. It assumes the scores are dense (not sparse features), and always in the same order. eg. if the 1st line has score a=2.1 b=3.1 c=0.3 then every other line must be a=??? b=??? c=???. I've attached the ini file that worked for me, and the output. I ran .../moses_chart -f moses.filtered.ini.1 -i input.split.1 >& out let me know if it works for you it'll be good if you can push your EMS changes when you're ready -- Hieu Hoang Research Associate University of Edinburgh http://www.hoang.co.uk/hieu
_______________________________________________ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
