Upload BertForJointParsing.py

BertForJointParsing.py

@@ -353,7 +353,6 @@ def convert_output_to_ud(output_sentences, htb_extras=False):
                 word['lex'] = word['seg'][-1]
             
             start = len(intermediate_output)
-            idx_to_key[word_idx] = len(intermediate_output) + 1
             # Add in all the prefixes
             if len(word['seg']) > 1:
                 for pre in get_prefixes_from_str(word['seg'][0], greedy=True):
@@ -370,6 +369,7 @@ def convert_output_to_ud(output_sentences, htb_extras=False):
                         intermediate_output[-1]['feats'] = 'Definite=Def|PronType=Art'
                     
                 
+            idx_to_key[word_idx] = len(intermediate_output) + 1
             # add the main word in!
             intermediate_output.append(dict(
                     word=word['seg'][-1], lex=word['lex'], pos=word['morph']['pos'], 
@@ -380,9 +380,9 @@ def convert_output_to_ud(output_sentences, htb_extras=False):
             if word['morph']['suffix']:
                 # first determine the dependency info:
                 # For adp, num, det - they main word points to here, and the suffix points to the dependency
+                entry_to_assign_suf_dep = None
                 if word['morph']['pos'] in ['ADP', 'NUM', 'DET']:
-                    intermediate_output[-1]['dep'] = len(intermediate_output)
-                    intermediate_output[-1]['absolute_dep'] = True
+                    entry_to_assign_suf_dep = intermediate_output[-1]
                     intermediate_output[-1]['func'] = 'case'
                     dep = word['syntax']['dep_head_idx']
                     func = word['syntax']['dep_func']
@@ -409,14 +409,18 @@ def convert_output_to_ud(output_sentences, htb_extras=False):
                 # for htb:
                 else:
                     # main word becomes the lexeme, the suffix is based on the features
-                    intermediate_output[-1]['word'] = s_lex + '_'
+                    intermediate_output[-1]['word'] = (s_lex if s_lex != s_word else s_word[:-1]) + '_'
                     suf_feats = word['morph']['suffix_feats']
-                    suf = ud_suffix_to_htb_str.get(f"Gender={suf_feats['Gender']}|Number={suf_feats['Number']}|Person={suf_feats['Person']}", "_הוא")
+                    suf = ud_suffix_to_htb_str.get(f"Gender={suf_feats.get('Gender', 'Fem,Masc')}|Number={suf_feats.get('Number', 'Sing')}|Person={suf_feats.get('Person', '3')}", "_הוא")
                     # for HTB, if the function is poss, then add a shel pointing to the next word
                     if func == 'nmod:poss':
-                        intermediate_output.append(dict(word='_של_', lex='של', pos='ADP', dep=len(intermediate_output) + 1, func='case', feats='_', absolute_dep=True))
+                        intermediate_output.append(dict(word='_של_', lex='של', pos='ADP', dep=len(intermediate_output) + 2, func='case', feats='_', absolute_dep=True))
                 # add the main suffix in
                 intermediate_output.append(dict(word=suf, lex='הוא', pos='PRON', dep=dep, func=func, feats='|'.join(f'{k}={v}' for k,v in word['morph']['suffix_feats'].items())))
+                if entry_to_assign_suf_dep:
+                    entry_to_assign_suf_dep['dep'] = len(intermediate_output)
+                    entry_to_assign_suf_dep['absolute_dep'] = True
+                    
             end = len(intermediate_output)
             ranges.append((start, end, word['token']))
             
@@ -458,12 +462,12 @@ def ud_get_prefix_dep(pre, word, word_idx):
         func = 'mark'
     # vuv goes to the main word if the function is in the list, otherwise follows
     elif pre == 'ו': 
-        does_follow_main = word['syntax']['dep_func'] in ["conj", "acl:recl", "parataxis", "root", "acl", "amod", "list", "appos", "dep", "flatccomp"]
+        does_follow_main = word['syntax']['dep_func'] not in ["conj", "acl:recl", "parataxis", "root", "acl", "amod", "list", "appos", "dep", "flatccomp"]
         func = 'cc'
     else:
         # for adj, noun, propn, pron, verb - prefixes go to the main word
         if word['morph']['pos'] in ["ADJ", "NOUN", "PROPN", "PRON", "VERB"]:
-            does_follow_main = True  
+            does_follow_main = False
         # otherwise - prefix follows the word if the function is in the list
         else: does_follow_main = word['syntax']['dep_func'] in ["compound:affix", "det", "aux", "nummod", "advmod", "dep", "cop", "mark", "fixed"]