On 25 Sep 2008, at 17:19, musa ghurab wrote: > Hi all > > i have 4 questions, I'm not clear enough about them > 1.I'm using the same corpus for the following steps: > 1.Build Language Model > 2.Train Model > 3.Tuning > 4.Run System on Development Test Set > 5.Evaluation > That is because i don't have another corpus to use for my language > pair, so is this OK or not? any suggestion about that?
This is not OK if you actually want to get a sense of how good your system works on *unseen* data. If you evaluate on data that was used to train the system, then the system has already essentially memorized the correct translations. This will make its accuracy appear artificially inflated. Normally you would use at least 3 separate data sets: Most of your data is used for steps 1-2. A small dataset (~1000 sentences) for step 3. A small dataset (~1000 sentences) for steps 4-5. > 2.What is the reason of using “Run System on Development Test Set” > on moses Baseline System at http://www.statmt.org/wmt08/baseline.html? > 3.When “Tuning” we get BLEU score, why should we evaluate the model > again “Evaluation” ? > 4.When Evaluating the model can i use source file and reference file > from the same corpus that i trained in my model? or should i provide > other than that? > 5.Is there any reference explains the mechanism of Tuning, > Evaluating, Weighting? Much of what you're asking is described on the Moses site: http://www.statmt.org/moses/ For a more general reference on statistical MT, I immodestly recommend the following article: http://homepages.inf.ed.ac.uk/alopez/papers/survey.pdf Cheers Adam -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
