Decoding Options

Travatar supports the following options during decoding. There is a possibility that this documenation is out of date, so you can confirm the latest options using travatar --help.

-config_file 	The location of the configuration file
-all_unk 	If this is true, translating the word as-is will be an option even when a rule exists
-binarize 	How to binarize the trees (none/left/right)
-debug 	What level of debugging output to print
-forest_out 	forest output file location
-in_format 	The format of the input (penn/egret)
-lm_file 	Language model file location
-nbest 	The length of the n-best list
-nbest_out 	n-best output file location
-pop_limit 	The number of pops necessary
-threads	The number of threads to use during decoding
-tm_file 	Translation model file location
-tm_storage 	Method of storing the rule table (marisa/hash)
-trace_out 	trace output file location
-weight_vals 	Weight values in format "name1=val1 name2=val2", existing features override the file, other features are left unchanged
-tune_update 	How to update the weights after each sentence is translated (none/perceptron)
-tune_ref_files 	The reference files to be used for tuning
-tune_step_count 	The number of steps that have been executed previously for tuning
-tune_step_size 	The size of a single step of weight update
-tune_loss 	The evaluation measure to use in tuning (bleu/ribes)
-tune_weight_out 	Location to print the weight file after done tuning