Added module responsible for sys uttr dialog history update

Signed-off-by: nvidia <tkornuta@nvidia.com>

examples/nlp/dialogue_state_tracking/rule_based_policy_multiwoz.py

@@ -41,6 +41,7 @@
 from nemo.collections.nlp.data.datasets.multiwoz_dataset.state import init_state
 from nemo.collections.nlp.nm.non_trainables import (
     RuleBasedDPMMultiWOZ,
+    SystemUtteranceHistoryUpdate,
     TemplateNLGMultiWOZ,
     TradeStateUpdateNM,
     UserUtteranceEncoder,
@@ -62,7 +63,7 @@
 ]
 
 
-def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state):
+def forward(dialog_pipeline, user_uttr, dial_history, belief_state):
     """
     Forward pass of the "Complete Dialog Pipeline".
 
@@ -75,7 +76,6 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
     
     Args:
         user_uttr (str): User utterance
-        system_uttr (str): Previous system utterance
         dialog_history (str): Dialogue history contains all previous system and user utterances
         belief_state (dict): dialogue state
     Returns:
@@ -87,7 +87,7 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
     # 1. Forward pass throught Word-Level Dialog State Tracking modules (TRADE).
     # 1.1. User utterance encoder.
     dialog_ids, dialog_lens, dial_history = dialog_pipeline.modules[dialog_pipeline.steps[0]].forward(
-        user_uttr=user_uttr, sys_uttr=system_uttr, dialog_history=dial_history,
+        user_uttr=user_uttr, dialog_history=dial_history,
     )
     # 1.2. TRADE encoder.
     outputs, hidden = dialog_pipeline.modules[dialog_pipeline.steps[1]].forward(
@@ -111,6 +111,11 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
     # 3. Forward pass throught Natural Language Generator module (Template-Based).
     system_uttr = dialog_pipeline.modules[dialog_pipeline.steps[5]].forward(system_acts=system_acts)
 
+    # 4. Update dialog  history with system utterance
+    dial_history = dialog_pipeline.modules[dialog_pipeline.steps[6]].forward(
+        sys_uttr=system_uttr, dialog_history=dial_history
+    )
+
     # Return the updated states and dialog history.
     return system_uttr, belief_state, dial_history
 
@@ -199,13 +204,16 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
     # NLG module.
     template_nlg = TemplateNLGMultiWOZ()
 
+    # Updates dialog history with system utterance.
+    sys_utter_history_update = SystemUtteranceHistoryUpdate()
+
     # Construct the "evaluation" (inference) neural graph by connecting the modules using nmTensors.
     # Note: Using the same names for passed nmTensor as in the actual forward pass.
     with NeuralGraph(operation_mode=OperationMode.evaluation) as dialog_pipeline:
         # 1.1. User utterance encoder.
         # Bind all the input ports of this module.
         dialog_ids, dialog_lens, dial_history = user_utterance_encoder(
-            user_uttr=dialog_pipeline, sys_uttr=dialog_pipeline, dialog_history=dialog_pipeline,
+            user_uttr=dialog_pipeline, dialog_history=dialog_pipeline,
         )
         # Fire step 1: 1.2. TRADE encoder.
         outputs, hidden = trade_encoder(inputs=dialog_ids, input_lens=dialog_lens)
@@ -229,6 +237,9 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
         # 3. Forward pass throught Natural Language Generator module (Template-Based).
         system_uttr = template_nlg(system_acts=system_acts)
 
+        # 4. Update dialog  history with system utterance
+        dial_history = sys_utter_history_update(sys_uttr=system_uttr, dialog_history=dial_history)
+
     # Show the graph summary.
     logging.info(dialog_pipeline.summary())
 
@@ -250,7 +261,7 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
             else:
                 # Pass the "user uterance" as inputs to the dialog pipeline.
                 system_uttr, belief_state, dial_history = forward(
-                    dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state
+                    dialog_pipeline, user_uttr, dial_history, belief_state
                 )
 
     elif args.mode == 'example':
@@ -262,5 +273,5 @@ def forward(dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state)
             for user_uttr in example:
                 logging.info("User utterance: %s", user_uttr)
                 system_uttr, belief_state, dial_history = forward(
-                    dialog_pipeline, system_uttr, user_uttr, dial_history, belief_state
+                    dialog_pipeline, user_uttr, dial_history, belief_state
                 )

nemo/collections/nlp/nm/non_trainables/__init__.py

@@ -15,6 +15,9 @@
 # =============================================================================
 
 from nemo.collections.nlp.nm.non_trainables.dialogue_state_tracking.rule_based_dpm_multiwoz import RuleBasedDPMMultiWOZ
+from nemo.collections.nlp.nm.non_trainables.dialogue_state_tracking.system_utterance_history_update import (
+    SystemUtteranceHistoryUpdate,
+)
 from nemo.collections.nlp.nm.non_trainables.dialogue_state_tracking.template_nlg_multiwoz import TemplateNLGMultiWOZ
 from nemo.collections.nlp.nm.non_trainables.dialogue_state_tracking.trade_state_update_nm import TradeStateUpdateNM
 from nemo.collections.nlp.nm.non_trainables.dialogue_state_tracking.user_utterance_encoder import UserUtteranceEncoder

nemo/collections/nlp/nm/non_trainables/dialogue_state_tracking/system_utterance_history_update.py

@@ -0,0 +1,88 @@
+# =============================================================================
+# Copyright 2020 NVIDIA. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# =============================================================================
+
+'''
+This file contains code artifacts adapted from the original implementation:
+https://github.com/thu-coai/ConvLab-2/
+'''
+import torch
+
+from nemo.backends.pytorch.nm import NonTrainableNM
+from nemo.core.neural_types import *
+from nemo.utils import logging
+from nemo.utils.decorators import add_port_docs
+
+__all__ = ['SystemUtteranceHistoryUpdate']
+
+
+class SystemUtteranceHistoryUpdate(NonTrainableNM):
+    """
+    Updates dialogue history with system utterance.
+    """
+
+    @property
+    @add_port_docs()
+    def input_ports(self):
+        """Returns definitions of module input ports.
+        user_uttr (str): user utterance
+        sys_uttr (str): system utterace
+        dialog_history (list): dialogue history, list of system and diaglogue utterances
+        """
+        return {
+            'sys_uttr': NeuralType(
+                axes=[AxisType(kind=AxisKind.Batch, is_list=True)], elements_type=SystemUtterance()
+            ),
+            'dialog_history': NeuralType(
+                axes=(AxisType(kind=AxisKind.Batch, is_list=True), AxisType(kind=AxisKind.Time, is_list=True),),
+                elements_type=AgentUtterance(),
+            ),
+        }
+
+    @property
+    @add_port_docs()
+    def output_ports(self):
+        """Returns definitions of module output ports.
+        dialog_history (list): dialogue history, being a list of user and system utterances.
+        """
+        return {
+            'dialog_history': NeuralType(
+                axes=(AxisType(kind=AxisKind.Batch, is_list=True), AxisType(kind=AxisKind.Time, is_list=True),),
+                elements_type=AgentUtterance(),
+            ),
+        }
+
+    def __init__(self):
+        """
+        Initializes the object
+        Args:
+            data_desc (obj): data descriptor for MultiWOZ dataset, contains information about domains, slots, 
+                    and associated vocabulary
+        """
+        super().__init__()
+
+    def forward(self, sys_uttr, dialog_history):
+        """
+        Returns updated dialog history.
+        Args:
+            sys_uttr (str): system utterace
+            dialog_history (list): dialogue history, list of user and system diaglogue utterances
+        Returns:
+            dialog_history (list): updated dialogue history, list of user and system diaglogue utterances
+        """
+        dialog_history.append(["sys", sys_uttr])
+        logging.debug("Dialogue history: %s", dialog_history)
+
+        return dialog_history

nemo/collections/nlp/nm/non_trainables/dialogue_state_tracking/template_nlg_multiwoz.py

@@ -111,7 +111,9 @@ def output_ports(self):
         system_uttr (str): generated system's response
         """
         return {
-            'sys_uttr': NeuralType(axes=[AxisType(kind=AxisKind.Batch, is_list=True)], elements_type=Utterance()),
+            'sys_uttr': NeuralType(
+                axes=[AxisType(kind=AxisKind.Batch, is_list=True)], elements_type=SystemUtterance()
+            ),
         }
 
     def forward(self, system_acts):

nemo/collections/nlp/nm/non_trainables/dialogue_state_tracking/user_utterance_encoder.py

@@ -30,7 +30,8 @@
 
 class UserUtteranceEncoder(NonTrainableNM):
     """
-    Encodes dialogue history (system and user utterances) into a Multiwoz dataset format
+    Updates dialogue history with user utterance and encodes the history (system and user utterances) into 
+    a flat list of tokens (per sample).
     Args:
         data_desc (obj): data descriptor for MultiWOZ dataset, contains information about domains, slots, 
             and associated vocabulary
@@ -41,12 +42,10 @@ class UserUtteranceEncoder(NonTrainableNM):
     def input_ports(self):
         """Returns definitions of module input ports.
         user_uttr (str): user utterance
-        sys_uttr (str): system utterace
         dialog_history (list): dialogue history, list of system and diaglogue utterances
         """
         return {
-            'user_uttr': NeuralType(axes=[AxisType(kind=AxisKind.Batch, is_list=True)], elements_type=Utterance()),
-            'sys_uttr': NeuralType(axes=[AxisType(kind=AxisKind.Batch, is_list=True)], elements_type=Utterance()),
+            'user_uttr': NeuralType(axes=[AxisType(kind=AxisKind.Batch, is_list=True)], elements_type=UserUtterance()),
             'dialog_history': NeuralType(
                 axes=(AxisType(kind=AxisKind.Batch, is_list=True), AxisType(kind=AxisKind.Time, is_list=True),),
                 elements_type=AgentUtterance(),
@@ -81,20 +80,18 @@ def __init__(self, data_desc):
         super().__init__()
         self.data_desc = data_desc
 
-    def forward(self, user_uttr, sys_uttr, dialog_history):
+    def forward(self, user_uttr, dialog_history):
         """
         Returns dialogue utterances in the format accepted by the TRADE Dialogue state tracking model
         Args:
             dialog_history (list): dialogue history, list of system and diaglogue utterances
             user_uttr (str): user utterance
-            sys_uttr (str): system utterace
         Returns:
             dialog_ids (int): token ids for the whole dialogue history
             dialog_lens (int): length of the whole tokenized dialogue history
             dialog_history (list): updated dialogue history, list of system and diaglogue utterances
         """
         # TODO: why we update sys utterance, whereas we have only user utterance at that point?
-        dialog_history.append(["sys", sys_uttr])
         dialog_history.append(["user", user_uttr])
         logging.debug("Dialogue history: %s", dialog_history)
 

nemo/core/neural_types/elements.py

@@ -44,6 +44,8 @@
     'StringLabel',
     'StringType',
     'Utterance',
+    'UserUtterance',
+    'SystemUtterance',
     'AgentUtterance',
     'TokenIndex',
     'Length',
@@ -258,14 +260,22 @@ class Utterance(StringType):
     """Element type representing an utterance (e.g. "Is there a train from Ely to Cambridge on Tuesday ?")."""
 
 
+class UserUtterance(Utterance):
+    """Element type representing a utterance expresesd by the user."""
+
+
+class SystemUtterance(Utterance):
+    """Element type representing an utterance produced by the system."""
+
+
 class AgentUtterance(ElementType):
     """Element type representing utterance returned by an agent (user or system) participating in a dialog."""
 
-    def __str__(self):
-        return "Utterance returned by an agent (user or system) participating in a dialog."
+    # def __str__(self):
+    #    return "Utterance returned by an agent (user or system) participating in a dialog."
 
-    def fields(self):
-        return ("agent", "utterance")
+    # def fields(self):
+    #    return ("agent", "utterance")
 
 
 class IntType(ElementType):