benchmark to colorlessgreen format

benchmark:
sentence_good
sentence_bad
one_prefix_prefix
one_prefix_word_good
one_prefix_word_bad
category
field
linguistics_term
UID
simple_LM_method
one_prefix_method
two_prefix_method
lexically_identical

colorlessgreen
pattern: just a code about something, maybe equivalent to linguistics_term or UID
constr_id: data-internal stuff, maybe if we number our construction type it'd be useful
sent_id: an id for each pair
correct_number: the label for the correct form
form: one_prefix_word_good/bad
class: good/bad
type: generated or original
prefix: one_prefix_prefix
n_attr: number of attractors
punct: does context include punctuation
freq: frequency of the form
len_context: ??
len_prefix
sent: 
