def analyse_operations_sentence(orig_sent, sys_sent, ref_sents, orig_parse,
sys_parse, ref_parses):
word_aligns_orig_sys = align(orig_parse, sys_parse)[0]
orig_annots, sys_annots = annotate_sentence(orig_sent.split(),
sys_sent.split(),
word_aligns_orig_sys,
orig_parse, sys_parse)
orig_auto_labels = _from_annots_to_labels(orig_annots, ORIG_OPS_LABELS,
'C')
curr_sent_scores = []
for ref_sent, ref_parse in zip(ref_sents, ref_parses):
word_aligns_orig_ref = align(orig_parse, ref_parse)[0]
orig_annots, ref_annots = annotate_sentence(orig_sent.split(),
ref_sent.split(),
word_aligns_orig_ref,
orig_parse, ref_parse)
orig_silver_labels = _from_annots_to_labels(orig_annots,
ORIG_OPS_LABELS, 'C')
f1_per_label = f1_score(orig_silver_labels,
orig_auto_labels,
labels=ORIG_OPS_LABELS,
average=None)
curr_sent_scores.append(f1_per_label)
return np.amax(curr_sent_scores, axis=0)
def test_align_str():
# aligning strings (output indexes start at 1)
sentence1 = "Four men died in an accident."
sentence2 = "4 people are dead from a collision."
alignments = aligner.align(sentence1, sentence2)
assert alignments[0] == [[7, 8], [2, 2], [3, 4], [1, 1], [6, 7], [5, 6]]
assert alignments[1] == [['.', '.'], ['men', 'people'], ['died', 'dead'], ['Four', '4'],
['accident', 'collision'], ['an', 'a']]
def test_align_tokens():
# aligning sets of tokens (output indexes start at 1)
sentence1 = ['Four', 'men', 'died', 'in', 'an', 'accident', '.']
sentence2 = ['4', 'people', 'are', 'dead', 'from', 'a', 'collision', '.']
alignments = aligner.align(sentence1, sentence2)
assert alignments[0] == [[7, 8], [2, 2], [3, 4], [1, 1], [6, 7], [5, 6]]
assert alignments[1] == [['.', '.'], ['men', 'people'], ['died', 'dead'], ['Four', '4'],
['accident', 'collision'], ['an', 'a']]
def align_scenes_sentences(scenes, synt_parse_sentences):
# parse the scenes
synt_parse_scenes = syntactic_parse_texts(scenes)
all_scenes_alignments = []
for synt_scene in synt_parse_scenes:
scene_alignments = []
for synt_sent in synt_parse_sentences:
# word_alignments = [[word1_scene, word1_sentence], [word2_scene, word3_sentence], ...]
word_alignments = align(synt_scene, synt_sent)[1]
scene_alignments.append(word_alignments)
all_scenes_alignments.append(scene_alignments)
return all_scenes_alignments
def align_scenes_sentences(synt_scenes, synt_sents, allow_mutiple_matches):
scenes_sents_aligns = []
already_matched = []
for synt_scene in synt_scenes:
max_sent_aligns = []
for sent_num, synt_sent in enumerate(synt_sents):
if not allow_mutiple_matches and sent_num in already_matched:
continue
word_alignments = align(synt_scene, synt_sent)[1]
if len(word_alignments) > len(max_sent_aligns):
max_sent_aligns = word_alignments
max_sent_num = sent_num
scenes_sents_aligns.append(max_sent_aligns)
if len(max_sent_aligns) > 0:
already_matched.append(max_sent_num)
return scenes_sents_aligns
