rikvn / amr Goto Github PK

Sometimes a produced AMR can have missing role but a colon:

(vv29traitor / traitor
    :quant (vv29many / many
        :degree (vv29too / too))
    : (vv29compatriot / compatriot
        :poss (vv29we / we)))

P.S. Despite this smatch is fine to translate them into triples that leads to obscure evaluation. smatch issue

Wiki links of the for X_(Y) are printed as X_ (Y), with an unwanted extra space.

For example: produced vs gold

(vv314vote-01 / vote-01 
	:ARG0 (vv314he / he) 
	:ARG1 (vv314bill / bill 
		:mod (vv314political-party / political-party :wiki "Democratic_Party_ (United_States) " 
			:name (v3141 / name :op1 "Dem")) 
		:quant (vv314several / several)) 
	:mod (vv314even / even))

# ::id DF-199-193694-586_5859.15 ::date 2013-08-29T08:01:13 ::annotator SDL-AMR-09 ::preferred
# ::snt He even voted for several Dem bills.
# ::save-date Mon Sep 16, 2013 ::file DF-199-193694-586_5859_15.txt
(v / vote-01
      :ARG0 (h / he)
      :ARG1 (b / bill
            :quant (s / several)
            :mod (p / political-party :wiki "Democratic_Party_(United_States)"
                  :name (n2 / name :op1 "Democratic" :op2 "Party")))
      :mod (e / even))

When I run the postprocess_AMRs.py as follows:
python postprocess_AMRs.py -f sample_alignment_input/sample.tf -s sample_alignment_input/sample.sent

I get the following error:
IOError: [Errno 2] No such file or directory: 'sample_alignment_input/sample.tf.restore'

I thought the postprocess file was supposed to create the sample.tf.restore file as well but it looks like it's not doing that. What am I missing here?

Note that here sample.tf refers to the predicted translation file from the OpenNMT model, and sample.sent is the actual sentence in English.

Hi,

Thanks for great project. In readme,

python postprocess_AMRs.py -f sample_alignment_input/sample.txt.char.tf -s sample_alignment_input/sample.sent

but there is no sample.txt.char.tf in sample_alignment_input?
Thanks in advance!