| def get_input_docs(doc_annot: dict): |
| """ InputDocs : list of lists with plain tokens of each document """ |
| return [token.text for token in doc_annot.get('tokens')] |
|
|
|
|
| def get_edu_breaks(doc_trees: list, doc_annot: dict): |
| """ EduBreak_TokenLevel : list of lists with the token positions of right EDU ends of each document """ |
|
|
| def extr_edus(tree, begin): |
| if tree.relation == 'elementary': |
| return [(tree.start - begin, tree.end - begin)] |
| else: |
| tt = [] |
| tt += extr_edus(tree.left, begin=begin) |
| tt += extr_edus(tree.right, begin=begin) |
| return tt |
|
|
| def map_offset_to_tokens(offset): |
| begin, end = -1, -1 |
| for i, token in enumerate(doc_annot['tokens']): |
| if begin == -1 and token.begin > offset[0]: |
| begin = i - 1 |
| if begin != -1: |
| if token.end > offset[1]: |
| end = i - 1 |
| return begin, end |
| return begin, i |
|
|
| edus = [] |
| for tree in doc_trees: |
| begin = tree.start |
| edus += extr_edus(tree, begin=tree.start) |
|
|
| return [map_offset_to_tokens(offset)[1] for offset in edus] |
|
|
|
|
| def get_sentence_breaks(doc_annot: dict): |
| """ SentBreak for sentence breaks in terms of token offsets """ |
| return [sentence.end - 1 for sentence in doc_annot.get('sentences')] |
|
|
|
|
| def leftmostid(tree): |
| if tree.left: |
| return leftmostid(tree.left) |
| return tree.id |
|
|
|
|
| def leftmoststart(tree): |
| if tree.left: |
| return leftmoststart(tree.left) |
| return tree.start |
|
|
|
|
| def rightmostid(tree): |
| if tree.right: |
| return rightmostid(tree.right) |
| return tree.id |
|
|
|
|
| def rightmostend(tree): |
| if tree.right: |
| return rightmostend(tree.right) |
| return tree.end |
|
|
|
|
| true_relations = ['attribution_NS', 'attribution_SN', 'background_NS', |
| 'cause-effect_NS', 'cause-effect_SN', |
| 'comparison_NN', 'concession_NS', 'condition_NS', 'condition_SN', |
| 'contrast_NN', 'elaboration_NS', 'evidence_NS', |
| 'interpretation-evaluation_NS', 'interpretation-evaluation_SN', |
| 'joint_NN', 'preparation_SN', 'purpose_NS', 'purpose_SN', |
| 'restatement_NN', 'same-unit_NN', 'sequence_NN', 'solutionhood_SN'] |
|
|
|
|
| def correct_relations(rel: str, nuc: str): |
| target_map = { |
| 'antithesis': 'attribution', |
| 'cause': 'cause-effect', |
| 'conclusion': 'restatement', |
| 'interpretation': 'interpretation-evaluation', |
| 'evaluation': 'interpretation-evaluation', |
| 'motivation': 'condition', |
| } |
|
|
| relation_map = { |
| 'effect_SN': 'cause-effect_NS', |
| 'effect_NS': 'cause-effect_SN', |
| 'evidence_SN': 'preparation_SN', |
| 'restatement_SN': 'condition_SN', |
| 'restatement_NS': 'elaboration_NS', |
| 'solutionhood_NS': 'elaboration_NS', |
| 'preparation_NS': 'elaboration_NS', |
| 'concession_SN': 'preparation_SN', |
| 'evaluation_SN': 'preparation_SN', |
| 'elaboration_SN': 'preparation_SN', |
| 'background_SN': 'preparation_SN', |
| } |
|
|
| if rel in target_map: |
| rel = target_map.get(rel) |
|
|
| full_rel = rel + '_' + nuc |
| if full_rel in relation_map: |
| full_rel = relation_map.get(full_rel) |
| rel, nuc = full_rel.split('_') |
|
|
| if not full_rel in true_relations: |
| print(rel, nuc, full_rel) |
|
|
| return rel, nuc |
|
|
|
|
| def du_to_docs_structure(tree: dict, du_counter: int, needs_preprocessing=True): |
| if tree.relation != 'elementary': |
| if needs_preprocessing: |
| tree.relation, tree.nuclearity = correct_relations(tree.relation, tree.nuclearity) |
|
|
| left_nuclearity = 'Satellite' if tree.nuclearity == 'SN' else 'Nucleus' |
| right_nuclearity = 'Satellite' if tree.nuclearity == 'NS' else 'Nucleus' |
| left_relation = tree.relation |
| right_relation = tree.relation |
|
|
| left_id_1 = leftmostid(tree.left) + du_counter |
| left_id_2 = rightmostid(tree.left) + du_counter |
| right_id_1 = leftmostid(tree.right) + du_counter |
| right_id_2 = rightmostid(tree.right) + du_counter |
|
|
| if left_nuclearity == 'Satellite': |
| right_relation = 'span' |
|
|
| if right_nuclearity == 'Satellite': |
| left_relation = 'span' |
|
|
| relstring_l = f'{left_id_1}:{left_nuclearity}={left_relation}:{left_id_2}' |
| relstring_r = f'{right_id_1}:{right_nuclearity}={right_relation}:{right_id_2}' |
|
|
| left_subtree_struct = du_to_docs_structure(tree.left, du_counter, needs_preprocessing) or [] |
| right_subtree_struct = du_to_docs_structure(tree.right, du_counter, needs_preprocessing) or [] |
| return [f'({relstring_l},{relstring_r})'] + left_subtree_struct + right_subtree_struct |
|
|
|
|
| def du_to_docs_structure_char(tree): |
| if tree.relation != 'elementary': |
| tree.relation, tree.nuclearity = correct_relations(tree.relation, tree.nuclearity) |
|
|
| left_nuclearity = 'Satellite' if tree.nuclearity == 'SN' else 'Nucleus' |
| right_nuclearity = 'Satellite' if tree.nuclearity == 'NS' else 'Nucleus' |
| left_relation = tree.relation |
| right_relation = tree.relation |
|
|
| left_id_1 = leftmoststart(tree.left) |
| left_id_2 = rightmostend(tree.left) |
| right_id_1 = leftmoststart(tree.right) |
| right_id_2 = rightmostend(tree.right) |
|
|
| if left_nuclearity == 'Satellite': |
| right_relation = 'span' |
|
|
| if right_nuclearity == 'Satellite': |
| left_relation = 'span' |
|
|
| relstring_l = f'{left_id_1}:{left_nuclearity}={left_relation}:{left_id_2}' |
| relstring_r = f'{right_id_1}:{right_nuclearity}={right_relation}:{right_id_2}' |
|
|
| left_subtree_struct = du_to_docs_structure_char(tree.left) or [] |
| right_subtree_struct = du_to_docs_structure_char(tree.right) or [] |
| return [f'({relstring_l},{relstring_r})'] + left_subtree_struct + right_subtree_struct |
|
|
|
|
| def charsonly(text): |
| return ''.join(text.replace('#', '').split()) |
|
|
|
|
| def du_to_docs_structure_charsonly(tree, text, needs_preprocessing=True): |
| if tree.relation != 'elementary': |
| if needs_preprocessing: |
| tree.relation, tree.nuclearity = correct_relations(tree.relation, tree.nuclearity) |
|
|
| left_nuclearity = 'Satellite' if tree.nuclearity == 'SN' else 'Nucleus' |
| right_nuclearity = 'Satellite' if tree.nuclearity == 'NS' else 'Nucleus' |
| left_relation = tree.relation |
| right_relation = tree.relation |
|
|
| left_id_1 = text.find(charsonly(tree.left.text)) |
| left_id_2 = left_id_1 + len(charsonly(tree.left.text)) |
| right_id_1 = text.find(charsonly(tree.right.text)) |
| right_id_2 = right_id_1 + len(charsonly(tree.right.text)) |
|
|
| if left_nuclearity == 'Satellite': |
| right_relation = 'span' |
|
|
| if right_nuclearity == 'Satellite': |
| left_relation = 'span' |
|
|
| relstring_l = f'{left_id_1}:{left_nuclearity}={left_relation}:{left_id_2}' |
| relstring_r = f'{right_id_1}:{right_nuclearity}={right_relation}:{right_id_2}' |
|
|
| left_subtree_struct = du_to_docs_structure_charsonly(tree.left, text, needs_preprocessing) or [] |
| right_subtree_struct = du_to_docs_structure_charsonly(tree.right, text, needs_preprocessing) or [] |
| return [f'({relstring_l},{relstring_r})'] + left_subtree_struct + right_subtree_struct |
|
|
|
|
| def collect_edus(docs_structure): |
| edus_id = [] |
| for entry in docs_structure: |
| left, right = entry.split(',') |
| left = left.replace('(', '').split(':') |
| du1, du2 = left[0], left[2] |
| if du1 == du2: |
| edus_id.append(int(du1)) |
|
|
| right = right.replace(')', '').split(':') |
| du1, du2 = right[0], right[2] |
| if du1 == du2: |
| edus_id.append(int(du1)) |
| return edus_id |
|
|
|
|
| def get_docs_structure(doc_trees: list): |
| result = [] |
| du_counter = 0 |
| for tree in doc_trees: |
| structure = du_to_docs_structure(tree, du_counter) |
| if structure: |
| result += structure |
| du_counter += len(structure) |
| else: |
| du_counter += 1 |
| return result |
|
|
|
|
| def get_docs_structure_charsonly(doc_trees: list, needs_preprocessing=True): |
| result = [] |
| text = charsonly(''.join([tree.text for tree in doc_trees])) |
| for tree in doc_trees: |
| structure = du_to_docs_structure_charsonly(tree, text, needs_preprocessing) |
| if structure: |
| result += structure |
| return result |
|
|
