melt2connlx: Convert MElt POS-tagger output into a CONLLX file

melt2connlx

#!/usr/bin/env ruby
#
# Convert `MElt -tdL` output into *.conllx format.  Usage:
#
#     melt2conllx < input.melt > output.conllx
#
# Input should be one line per sentence, formatted like:
#
#     Durant/P/durant le/DET/le trajet/NC/trajet qui/PROREL/qui
#
# The first field is the surface form, the second is the POSTAG, and the
# third is the lemma.

# Generate course-grained part-of-speech tags from regular part-of-speech
# tags, in hopes that it will help the parser.
CPOSTAG = {
  "CLS" => "CL",
  "V" => "V",
  "DET" => "D",
  "NC" => "N",
  "CC" => "C",
  "ADJ" => "A",
  "DET" => "D",
  "NC" => "N",
  "P" => "P",
  "ADV" => "ADV",
  "VINF" => "V",
  "CLS" => "CL",
  "CLO" => "CL",
  "PONCT" => "PONCT",
  "ADJWH" => "A",
  "ADVWH" => "A",
  "CLR" => "CL",
  "CS" => "C",
  "DETWH" => "D",
  "ET" => "ET",
  "I" => "I",
  "NPP" => "N",
  "P+D" => "P+D",
  "P+PRO" => "P+PRO",
  "PREF" => "PREF",
  "PRO" => "PRO",
  "PROREL" => "PRO",
  "PROWH" => "PRO",
  "VIMP" => "V",
  "VPP" => "V",
  "VPR" => "V",
  "VS" => "V"
}

# One sentence per line.
STDIN.each_line do |line|
  # Convert each word into a CONLLX row.
  line.chomp.split(/ /).each_with_index do |token, index|
    form_and_postag, slash, lemma = token.rpartition('/')
    form, slash, postag = form_and_postag.rpartition('/')
    cpostag = CPOSTAG[postag] || postag
    puts([index+1, form, lemma, cpostag, postag, "_", "_"].join("\t"))
  end

  # Separate sentences with a blank line, or else maltparser will get
  # confused.
  puts
end