Warning: No built-in rules for language de.

[shiny1022]

I‘m training a tensorflow-based Nematus system. When I translate English into German, there are some problems.
here is the script:

src=en
trg=de
device=0
model=$working_dir/model.best-valid-script
dev_prefix=dev
dev=$dev_prefix.bpe.$src
ref=$dev_prefix.$trg

# decode
CUDA_VISIBLE_DEVICES=$device python $nematus/nematus/translate.py \
     -m $model \
     -i $data_dir/$dev \
     -o $working_dir/$dev.output.dev \
     -k 12 \
     -n
# evaluate translations and write BLEU score to standard output (for use by nmt.py)
$nematus/postprocess.sh < $working_dir/$dev.output.dev | \
$nematus/data/multi-bleu-detok.perl data/$ref | \
    cut -f 3 -d ' ' | \
    cut -f 1 -d ','

I got this error：

Warning: No built-in rules for language de.
Detokenizer Version $Revision: 4134 $
Language: de
Use of uninitialized value $length_reference in numeric eq (==) at /home/moses/NLP/workspace/mt/data/multi-bleu-detok.perl line 155.
0

Anyone can help me ?
Thanks!

[pjwilliams]

Does the translation output file ( $working_dir/$dev.output.dev ) contain any translations? When I see that unintialized $length_reference error it usually means the translate.py command failed resulting in an empty output, which the BLEU script has then tried (and failed) to score.

…

On 10 Dec 2018, at 13:26, shiny1022 ***@***.***> wrote: I‘m training a tensorflow-based Nematus system. When I translate English into German, there are some problems. here is the script: src=en trg=de device=0 model=$working_dir/model.best-valid-script dev_prefix=dev dev=$dev_prefix.bpe.$src ref=$dev_prefix.$trg decode CUDA_VISIBLE_DEVICES=$device python $nematus/nematus/translate.py -m $model -i $data_dir/$dev -o $working_dir/$dev.output.dev -k 12 -n evaluate translations and write BLEU score to standard output (for use by nmt.py) $nematus/postprocess.sh < $working_dir/$dev.output.dev | $nematus/data/multi-bleu-detok.perl data/$ref | cut -f 3 -d ' ' | cut -f 1 -d ',' I got this error： Warning: No built-in rules for language de. Detokenizer Version $Revision: 4134 $ Language: de Use of uninitialized value $length_reference in numeric eq (==) at /home/moses/NLP/workspace/mt/data/multi-bleu-detok.perl line 155. 0 Anyone can help me ? Thanks! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#92>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABDaYwX1L5s73os2JxA_XdSjLll7BDSFks5u3mEOgaJpZM4ZLW_d>.

[shiny1022]

Yes，the translation output file is empty. Then I ran the translate.py and it still couldn't work and occured the same error. No config files missing. It seemed that the model could‘t translate the input file.Here may be some problems in the process of translation.I'm confused about that.

[bricksdont]

Can you please provide a complete script, where all variables are defined? Examples of missing variables: $working_dir, $nematus.

Please check the contents of the file $data_dir/$dev. How many lines are in this file?

Which version of Nematus are you using, i.e. which commit?