Merge pull request #130 from lenka-krippnerova/master

Improve handling of negation detection in Slavic blocks

Author: dan-zeman

udapi/block/msf/phrase.py

@@ -30,6 +30,31 @@ def process_node(self, node):
 		'animacy':'PhraseAnimacy',
 		'ords':'Phrase'
 		}
+    
+    # a dictionary where the key is the lemma of a negative particle and the value is a list of the lemmas of their possible children that have a 'fixed' relation
+    # we do not want to include these negative particles in the phrase; these are expressions like "never", etc.
+    negation_fixed = {
+        # Belarusian
+        'ні' : ['раз'],
+        'ня' : ['толькі'],
+
+        # Upper Sorbian
+        'nic' : ['naposledku'],
+
+        # Polish
+        'nie' : ['mało'],
+
+        # Pomak
+        'néma' : ['kak'],
+
+        # Slovenian
+        'ne' : ['le'],
+
+        # Russian and Old East Slavic
+        'не' : ['то', 'токмо'],
+        'ни' : ['в', 'раз', 'шатко'],
+        'нет' : ['нет']
+    }
 
     def write_node_info(self, node,
 			tense = None,
@@ -51,12 +76,46 @@ def write_node_info(self, node,
             if val != None: 
                 node.misc[self.dictionary[key]] = val
 
-    def get_polarity(self, node, neg):
-        if node.feats['Polarity'] != "":
-            return node.feats['Polarity']
-        if len(neg) == 0:
-            return None
-        return 'Neg'
+    def has_fixed_children(self, node):
+        """
+        Returns True if the node has any children with the 'fixed' relation and the node's lemma along with the child's lemma are listed in self.negation_fixed.
+        """
+        fixed_children = [x for x in node.children if x.udeprel == 'fixed']
+
+        if fixed_children:
+            if fixed_children[0].lemma in self.negation_fixed.get(node.lemma, []):
+                return True
+        return False
+
+    def get_polarity(self, nodes):
+        """
+        Returns 'Neg' if there is exactly one node with Polarity='Neg' among the given nodes.
+        Returns an empty string if there are zero or more than one such nodes.
+        """
+        neg_count = 0
+        for node in nodes:
+            if node.feats['Polarity'] == 'Neg':
+                neg_count += 1
+
+        if neg_count == 1:
+            return 'Neg'
+        
+        # neg_count can be zero or two, in either case we want to return an empty string so that the PhrasePolarity attribute is not generated
+        else:
+            return ''
+               
+    def get_negative_particles(self, nodes):
+        """
+        Returns a list of all negative particles found among the children 
+        of the specified nodes, except for negative particles with fixed children specified in self.negation_fixed.
+        """
+        neg_particles = []
+        for node in nodes:
+            neg = [x for x in node.children if x.upos == 'PART' and x.feats['Polarity'] == 'Neg' and x.udeprel == 'advmod' and not self.has_fixed_children(x)]
+            if neg:
+                neg_particles += neg
+        return neg_particles
+            
 
     def get_is_reflex(self,node,refl):
         if node.feats['Voice'] == 'Mid':
@@ -75,4 +134,4 @@ def get_voice(self,node,refl):
         if self.is_expl_pass(refl):
             return 'Pass'
         return voice
-		
+

udapi/block/msf/slavic/conditional.py

@@ -19,12 +19,16 @@ def process_node(self, node):
 			# the conditional mood can be formed using the auxiliary verb or some conjunctions (such as 'aby, kdyby...' in Czech)
 			# so x.udeprel == 'aux' can't be required because it doesn't meet the conjunctions
 
-			if len(aux_cnd) > 0 and len(cop) == 0:
+			if aux_cnd and not cop:
 				aux = [x for x in node.children if x.udeprel == 'aux' or x.feats['Mood'] == 'Cnd'] # all auxiliary verbs and conjuctions with feats['Mood'] == 'Cnd'
 				refl = [x for x in node.children if x.feats['Reflex'] == 'Yes' and x.udeprel == 'expl']
-				neg = [x for x in node.children if x.feats['Polarity'] == 'Neg' and x.upos == 'PART']
+
+				phrase_nodes = [node] + aux + refl
 
-				phrase_ords = [node.ord] + [x.ord for x in aux] + [x.ord for x in refl] + [x.ord for x in neg]
+				neg = self.get_negative_particles(phrase_nodes)
+				phrase_nodes += neg
+
+				phrase_ords = [x.ord for x in phrase_nodes]
 				phrase_ords.sort()
 
 				auxVerb = aux_cnd[0]
@@ -41,7 +45,7 @@ def process_node(self, node):
 					form='Fin',
 					aspect=node.feats['Aspect'],
 					reflex=self.get_is_reflex(node,refl),
-					polarity=self.get_polarity(node,neg),
+					polarity=self.get_polarity(phrase_nodes),
 					voice=self.get_voice(node, refl),
 					ords=phrase_ords,
 					gender=node.feats['Gender'],
@@ -53,15 +57,18 @@ def process_node(self, node):
 		cop = [x for x in node.children if x.udeprel == 'cop' and (x.feats['VerbForm'] == 'Part' or x.feats['VerbForm'] == 'Fin')]
 		aux_cnd = [x for x in node.children if x.feats['Mood'] == 'Cnd' or x.deprel=='aux:cnd']
 
-		if len(cop) > 0 and len(aux_cnd) > 0:
+		if cop and aux_cnd:
 			# there can be a copula with Mood='Cnd' (i. e. in Old East Slavonic), we don't want to count these copula in phrase_ords twice, so there is x.udeprel != 'cop' in aux list
 			aux = [x for x in node.children if (x.udeprel == 'aux' or x.feats['Mood'] == 'Cnd') and x.udeprel != 'cop']
-			neg = [x for x in node.children if x.feats['Polarity'] == 'Neg' and x.upos == 'PART']
 			prep = [x for x in node.children if x.upos == 'ADP']
 			refl = [x for x in node.children if x.feats['Reflex'] == 'Yes' and x.udeprel == 'expl']
 
+			phrase_nodes = [node] + aux + prep + refl + cop
+			neg = self.get_negative_particles(phrase_nodes)
+			phrase_nodes += neg
+
 			copVerb = cop[0]
-			phrase_ords = [node.ord] + [x.ord for x in aux] + [x.ord for x in cop] + [x.ord for x in neg] + [x.ord for x in prep] + [x.ord for x in refl]
+			phrase_ords = [x.ord for x in phrase_nodes]
 			phrase_ords.sort()
 			self.write_node_info(node,
 					aspect=copVerb.feats['Aspect'],
@@ -70,9 +77,9 @@ def process_node(self, node):
 					mood='Cnd',
 					form='Fin',
 					voice=self.get_voice(copVerb, refl),
-					polarity=self.get_polarity(copVerb,neg),
+					polarity=self.get_polarity(phrase_nodes),
 					reflex=self.get_is_reflex(node, refl),
 					ords=phrase_ords,
 					gender=copVerb.feats['Gender'],
 					animacy=copVerb.feats['Animacy']
-			)
+			)

udapi/block/msf/slavic/converb.py

@@ -12,9 +12,12 @@ def process_node(self, node):
 		# condition node.upos == 'VERB' to prevent copulas from entering this branch
 		if node.feats['VerbForm'] == 'Conv' and node.upos == 'VERB':
 			refl = [x for x in node.children if x.feats['Reflex'] == 'Yes' and x.udeprel == 'expl']
-			neg = [x for x in node.children if x.feats['Polarity'] == 'Neg' and x.upos == 'PART']
-			
-			phrase_ords = [node.ord] + [x.ord for x in refl] + [x.ord for x in neg]
+
+			phrase_nodes = [node] + refl
+			neg = self.get_negative_particles(phrase_nodes)
+			phrase_nodes += neg
+
+			phrase_ords = [x.ord for x in phrase_nodes]
 			phrase_ords.sort()
 
 			self.write_node_info(node,
@@ -23,7 +26,7 @@ def process_node(self, node):
 				form='Conv',
 				tense=node.feats['Tense'],
 				aspect=node.feats['Aspect'],
-				polarity=self.get_polarity(node,neg),
+				polarity=self.get_polarity(phrase_nodes),
 				reflex=self.get_is_reflex(node,refl),
 				ords=phrase_ords,
 				gender=node.feats['Gender'],
@@ -35,10 +38,13 @@ def process_node(self, node):
 		elif node.upos == 'ADJ':
 			aux = [x for x in node.children if x.udeprel == 'aux' and x.feats['VerbForm'] == 'Conv']
 
-			if len(aux) > 0:
-				neg = [x for x in node.children if x.feats['Polarity'] == 'Neg' and x.upos == 'PART']
+			if aux:
 				auxVerb = aux[0]
-				phrase_ords = [node.ord] + [x.ord for x in aux] + [x.ord for x in neg]
+
+				phrase_nodes = [node] + aux
+				neg = self.get_negative_particles(phrase_nodes)
+				phrase_nodes += neg
+				phrase_ords = [x.ord for x in phrase_nodes]
 				phrase_ords.sort()
 
 				self.write_node_info(node,
@@ -47,7 +53,7 @@ def process_node(self, node):
 					form='Conv',
 					tense=auxVerb.feats['Tense'],
 					aspect=node.feats['Aspect'],
-					polarity=self.get_polarity(auxVerb,neg),
+					polarity=self.get_polarity(phrase_nodes),
 					ords=phrase_ords,
 					gender=auxVerb.feats['Gender'],
 					animacy=auxVerb.feats['Animacy'],
@@ -58,13 +64,16 @@ def process_node(self, node):
 		else:
 			cop = [x for x in node.children if x.udeprel == 'cop' and x.feats['VerbForm'] == 'Conv']
 
-			if len(cop) > 0:
+			if cop:
 				prep = [x for x in node.children if x.upos == 'ADP']
-				neg = [x for x in node.children if x.feats['Polarity'] == 'Neg' and x.upos == 'PART']
 				refl = [x for x in node.children if x.feats['Reflex'] == 'Yes' and x.udeprel == 'expl']
 
 				copVerb = cop[0]
-				phrase_ords = [node.ord] + [x.ord for x in cop] + [x.ord for x in prep] + [x.ord for x in neg] + [x.ord for x in refl]
+
+				phrase_nodes = [node] + cop + prep + refl
+				neg = self.get_negative_particles(phrase_nodes)
+				phrase_nodes += neg
+				phrase_ords = [x.ord for x in phrase_nodes]
 				phrase_ords.sort()
 
 
@@ -76,7 +85,7 @@ def process_node(self, node):
 					gender=copVerb.feats['Gender'],
 					animacy=copVerb.feats['Animacy'],
 					form='Conv',
-					polarity=self.get_polarity(node,neg),
+					polarity=self.get_polarity(phrase_nodes),
 					ords=phrase_ords,
 					voice=self.get_voice(copVerb, refl)
 					)