Improve question detection

modules/ravestate_nlp/__init__.py

@@ -42,7 +42,7 @@ def roboy_getter(doc) -> bool:
 with rs.Module(name="nlp"):
 
     prop_tokens = rs.Property(name="tokens", default_value="", always_signal_changed=True, allow_pop=False, allow_push=False)
-    porp_postags = rs.Property(name="postags", default_value="", always_signal_changed=True, allow_pop=False, allow_push=False)
+    prop_postags = rs.Property(name="postags", default_value="", always_signal_changed=True, allow_pop=False, allow_push=False)
     prop_lemmas = rs.Property(name="lemmas", default_value="", always_signal_changed=True, allow_pop=False, allow_push=False)
     prop_tags = rs.Property(name="tags", default_value="", always_signal_changed=True, allow_pop=False, allow_push=False)
     prop_ner = rs.Property(name="ner", default_value="", always_signal_changed=True, allow_pop=False, allow_push=False)
@@ -55,7 +55,7 @@ def roboy_getter(doc) -> bool:
     sig_intent_play = rs.Signal(name="intent-play")
 
 
-    @rs.state(read=rawio.prop_in, write=(prop_tokens, porp_postags, prop_lemmas, prop_tags, prop_ner, prop_triples, prop_roboy, prop_yesno))
+    @rs.state(read=rawio.prop_in, write=(prop_tokens, prop_postags, prop_lemmas, prop_tags, prop_ner, prop_triples, prop_roboy, prop_yesno))
     def nlp_preprocess(ctx):
         text = ctx[rawio.prop_in]
         if not text:
@@ -68,7 +68,7 @@ def nlp_preprocess(ctx):
         logger.info(f"[NLP:tokens]: {nlp_tokens}")
 
         nlp_postags = tuple(str(token.pos_) for token in nlp_doc)
-        ctx[porp_postags] = nlp_postags
+        ctx[prop_postags] = nlp_postags
         logger.info(f"[NLP:postags]: {nlp_postags}")
 
         nlp_lemmas = tuple(str(token.lemma_) for token in nlp_doc)
@@ -84,6 +84,7 @@ def nlp_preprocess(ctx):
         logger.info(f"[NLP:ner]: {nlp_ner}")
 
         nlp_triples = nlp_doc._.triples
+        nlp_triples[0].set_yesno_question(detect_yesno_question(nlp_postags))
         ctx[prop_triples] = nlp_triples
         logger.info(f"[NLP:triples]: {nlp_triples}")
 
@@ -101,7 +102,7 @@ def nlp_contains_roboy_signal(ctx):
             return rs.Emit()
         return False
 
-    @rs.state(signal=sig_is_question, read=prop_triples)
+    @rs.state(signal=sig_is_question, read=(prop_triples, prop_tags))
     def nlp_is_question_signal(ctx):
         if ctx[prop_triples][0].is_question():
             return rs.Emit()
@@ -114,3 +115,9 @@ def nlp_intent_play_signal(ctx):
             return rs.Emit()
         return False
 
+    def detect_yesno_question(tags):
+        """
+        tests whether the prop_tags indicate that a yesno-question was asked
+        """
+        return tags[0] in {'VBP', 'VBD', 'VBZ', 'MD'} and tags[1] in {'PRP', 'DT'} or \
+            tags[0] in {'VBP', 'VBD', 'VBZ', 'MD'} and tags[1] == 'RB' and tags[2] in {'PRP', 'DT'}

modules/ravestate_nlp/triple.py

@@ -9,6 +9,7 @@ class Triple:
     _predicate: Token
     _predicate_aux: Token
     _object: Union[Token, QuestionWord]
+    _yesno_question: bool
 
     def __init__(self, subject: Token = None, predicate: Token = None, predicate_aux: Token=None, object: Token = None):
         self.set_subject(subject)
@@ -28,6 +29,9 @@ def set_predicate_aux(self, predicate_aux: Token):
     def set_object(self, object: Union[Token, QuestionWord]):
         self._object = object
 
+    def set_yesno_question(self, is_yesno: bool):
+        self._yesno_question = is_yesno
+
     def get_subject(self) -> Union[Token, QuestionWord]:
         return self._subject
 
@@ -40,6 +44,9 @@ def get_predicate_aux(self) -> Token:
     def get_object(self) -> Union[Token, QuestionWord]:
         return self._object
 
+    def get_yesno_question(self) -> bool:
+        return self._yesno_question
+
     def has_subject(self) -> bool:
         return self._subject and len(self._subject.text.strip()) > 0
 
@@ -79,7 +86,9 @@ def is_question(self, question_word: Optional[str] = None):
         if question_word:
             return self.match_either_lemma(obj={question_word}, subj={question_word})
         else:
-            return isinstance(self._subject, QuestionWord) or isinstance(self._object, QuestionWord)
+            return self._yesno_question \
+                   or isinstance(self._subject, QuestionWord) \
+                   or isinstance(self._object, QuestionWord)
 
     def ensure_notnull(self, empty_token):
         # do not allow empty entries in triple

test/modules/ravestate_nlp/test_preprocessing.py

@@ -1,5 +1,5 @@
 import pytest
-from ravestate_nlp import nlp_preprocess, prop_tokens, porp_postags, prop_lemmas, prop_tags, prop_ner, prop_roboy
+from ravestate_nlp import nlp_preprocess, prop_tokens, prop_postags, prop_lemmas, prop_tags, prop_ner, prop_roboy
 from ravestate_rawio import prop_in as raw_in
 from testfixtures import log_capture
 
@@ -23,7 +23,7 @@ def test_tokenization(capture, basic_input):
 def test_postags(capture, basic_input):
     nlp_preprocess(basic_input)
     expected = ('INTJ', 'NOUN', 'DET', 'NOUN', 'VERB', 'ADJ')
-    assert basic_input[porp_postags] == expected
+    assert basic_input[prop_postags] == expected
     capture.check_present((f"{FILE_NAME}", 'INFO', f"[NLP:postags]: {expected}"))