#!/usr/bin/perl
#
# Convert interest.acl94.txt into interest.arff
# [Skeleton for code, showing most of the relevant "gotchas"]

my @lines = <STDIN>;   # Slurp in entire file; use only for small files!
my $linenum = 0;

foreach my $line (@lines) {

    # Ignore $$ separator lines
    next if ($line =~ /^\s*\$\$\s*$/);

    my @words;
    my @tags;


    # Clean up line formatting, e.g. get rid of brackets, ====
    ...

    # Convert plural to singular for interest(s), lose underscore in sense labels
    # since WEKA doesn't like punctuation in its data items
    # e.g. interests_6 becomes interest6
    ...

    # Get word/tag pairs
    for ($i = 0; $i < @pairs; ++$i)
    {
        ...

        # Replace all punctuation within values with X, since WEKA doesn't like punctuation
        # (e.g. u.s. becomes uXsX)
        # Except %, which turns into PERCENT (worth keeping for this problem!)

        ...

        # (For optional experiment with coarser sense distinctions)
        # Collapse interest1...interest4   to interestA (interest=caring about)
        # Collapse interest5 and interest6 to interestB (interest=financial)

        ...
    }

    # Go through this sentence
    for ($i = 0; $i < @pairs; ++$i)
    {
        if ($words[$i] =~ /^interest[0-9AB]/)
        {
            # This is the target word position.  Record features around it for this sentence.
            # Let prevword, prevtag = NULL if interest is first word in sentence
            # Let nextword, nexttag = NULL if interest is last  word in sentence

            ...
        }
        else
        {
            # For other non-target words, just record the word and tag as having occurred,
            # since we will need to include them in the lists of values for this nominal
            # @attribute in the header.

            ...
        }
    }

    ++$linenum;
}

# Build nominal attribute value lists containing recorded words (for prevword,nextword)
# and tags (for prevtag, nexttag)

...

# Output .arff format (header, then one line of attributes and class for each sentence )


...