Implementing Schema: End-to-end Task-oriented Dialogue System (WoZ)

seacrowd/utils/constants.py

@@ -2,14 +2,25 @@
 from enum import Enum
 from types import SimpleNamespace
 
-from seacrowd.utils.schemas import (image_text_features, kb_features,
-                                    pairs_features, pairs_features_score,
-                                    pairs_multi_features, qa_features,
-                                    seq_label_features, speech2speech_features,
-                                    speech_features, speech_multi_features,
-                                    speech_text_features, ssp_features,
-                                    text2text_features, text_features,
-                                    text_multi_features, video_features)
+from seacrowd.utils.schemas import (
+    image_text_features,
+    kb_features,
+    pairs_features,
+    pairs_features_score,
+    pairs_multi_features,
+    qa_features,
+    seq_label_features,
+    speech2speech_features,
+    speech_features,
+    speech_multi_features,
+    speech_text_features,
+    ssp_features,
+    text2text_features,
+    text_features,
+    text_multi_features,
+    video_features,
+    tod_features,
+)
 
 METADATA: dict = {
     "_LOCAL": bool,
@@ -74,6 +85,7 @@ class Tasks(Enum):
 
     # Multi Text Generation
     DIALOGUE_SYSTEM = "DS"
+    E2E_TASK_ORIENTED_DIALOGUE = "TOD"
 
     # Self Supervised & Unsupervised Text
     PROMPTING = "PRT"
@@ -194,34 +206,28 @@ class Licenses(Enum):
     Tasks.WORD_ANALOGY: "T2T",
     Tasks.KEYWORD_EXTRACTION: "SEQ_LABEL",
     Tasks.COREFERENCE_RESOLUTION: "KB",
-
     Tasks.DIALOGUE_SYSTEM: "T2T",
-
+    Tasks.E2E_TASK_ORIENTED_DIALOGUE: "TOD",
     Tasks.NAMED_ENTITY_RECOGNITION: "SEQ_LABEL",
     Tasks.POS_TAGGING: "SEQ_LABEL",
     Tasks.KEYWORD_TAGGING: "SEQ_LABEL",
     Tasks.SENTENCE_ORDERING: "SEQ_LABEL",
     Tasks.TOKEN_LEVEL_LANGUAGE_IDENTIFICATION: "SEQ_LABEL",
-
     Tasks.COMMONSENSE_REASONING: "QA",
     Tasks.QUESTION_ANSWERING: "QA",
-
     Tasks.NEXT_SENTENCE_PREDICTION: "PAIRS",
     Tasks.TEXT_RETRIEVAL: "PAIRS",
     Tasks.TEXTUAL_ENTAILMENT: "PAIRS",
     Tasks.SEMANTIC_SIMILARITY: "PAIRS_SCORE",
     Tasks.SHORT_ANSWER_GRADING: "PAIRS_SCORE",
     Tasks.MORPHOLOGICAL_INFLECTION: "PAIRS_MULTI",
-
     Tasks.INSTRUCTION_TUNING: "T2T",
     Tasks.PARAPHRASING: "T2T",
     Tasks.MACHINE_TRANSLATION: "T2T",
     Tasks.SUMMARIZATION: "T2T",
     Tasks.MULTILEXNORM: "T2T",
     Tasks.TRANSLITERATION: "T2T",
-
     Tasks.ASPECT_BASED_SENTIMENT_ANALYSIS: "TEXT_MULTI",
-
     Tasks.SENTIMENT_ANALYSIS: "TEXT",
     Tasks.TAX_COURT_VERDICT: "TEXT",
     Tasks.EMOTION_CLASSIFICATION: "TEXT",
@@ -230,32 +236,24 @@ class Licenses(Enum):
     Tasks.INTENT_CLASSIFICATION: "TEXT",
     Tasks.RHETORIC_MODE_CLASSIFICATION: "TEXT",
     Tasks.TOPIC_MODELING: "TEXT",
-
     Tasks.PROMPTING: "SSP",
     Tasks.SELF_SUPERVISED_PRETRAINING: "SSP",
-
     Tasks.SPEECH_RECOGNITION: "SPTEXT",
     Tasks.SPEECH_TO_TEXT_TRANSLATION: "SPTEXT",
     Tasks.TEXT_TO_SPEECH: "SPTEXT",
-
     Tasks.SPEECH_TO_SPEECH_TRANSLATION: "S2S",
-
     Tasks.SPEECH_LANGUAGE_IDENTIFICATION: "SPEECH",
     Tasks.SPEECH_EMOTION_RECOGNITION: "SPEECH",
-
     Tasks.SPEECH_EMOTION_RECOGNITION_MULTILABEL: "SPEECH_MULTI",
-
     Tasks.IMAGE_CAPTIONING: "IMTEXT",
     Tasks.STYLIZED_IMAGE_CAPTIONING: "IMTEXT",
     Tasks.VISUALLY_GROUNDED_REASONING: "IMTEXT",
-
     Tasks.HOAX_NEWS_CLASSIFICATION: "TEXT",
     Tasks.CONCEPT_ALIGNMENT_CLASSIFICATION: "PAIRS",
     Tasks.SPAN_BASED_ABSA: "SEQ_LABEL",
     Tasks.FACT_CHECKING: None,
     Tasks.VIDEO_CAPTIONING: "VIDTEXT",
     Tasks.VIDEO_TO_TEXT_RETRIEVAL: "VIDTEXT",
-
     Tasks.CONCEPT_ALIGNMENT_CLASSIFICATION: "PAIRS",
     Tasks.FACT_CHECKING: None,
 }
@@ -285,6 +283,7 @@ class Licenses(Enum):
     "SPEECH_MULTI": speech_multi_features(),
     "IMTEXT": image_text_features(),
     "VIDTEXT": video_features,
+    "TOD": tod_features,
 }
 
 TASK_TO_FEATURES = {

seacrowd/utils/schemas/__init__.py

@@ -14,6 +14,7 @@
 from .text_multilabel import features as text_multi_features
 from .text_to_text import features as text2text_features
 from .video import features as video_features
+from .tod import features as tod_features
 
 __all__ = [
     "image_text_features",
@@ -32,4 +33,5 @@
     "text_multi_features",
     "text2text_features",
     "video_features",
+    "tod_features",
 ]

seacrowd/utils/schemas/tod.py

@@ -0,0 +1,74 @@
+"""
+End-to-end Task Oriented Dialogue Schema
+
+This schema will adapt the WoZ schema that has been used in several end-to-end task-oriented dialogue (TOD) task.
+URL: https://huggingface.co/datasets/woz_dialogue
+"""
+import datasets
+
+"""
+EXAMPLE:
+
+[
+	{
+		"turn_label": [ [ "food", "eritrean" ] ],
+		"system_utterance": "",
+		"turn_idx": 0,
+		"belief_state": [ { "slots": [ [ "food", "eritrean" ] ], "act": "inform" } ],
+		"user_utterance": "Are there any eritrean restaurants in town?",
+		"system_acts": []
+	},
+	{
+		"turn_label": [ [ "food", "chinese" ] ],
+		"system_utterance": "No, there are no eritrean restaurants in town. Would you like a different restaurant? ",
+		"turn_idx": 1,
+		"belief_state": [ { "slots": [ [ "food", "chinese" ] ], "act": "inform" } ],
+		"user_utterance": "How about Chinese food?",
+		"system_acts": []
+	},
+	{
+		"turn_label": [ [ "area", "east" ] ],
+		"system_utterance": "There is a wide variety of Chinese restaurants, do you have an area preference or a price preference to narrow it down?",
+		"turn_idx": 2,
+		"belief_state": [ { "slots": [ [ "food", "chinese" ] ], "act": "inform" }, { "slots": [ [ "area", "east" ] ], "act": "inform" } ],
+		"user_utterance": "I would like the East part of town.",
+		"system_acts": [ [ "area" ] ]
+	},
+	{
+		"turn_label": [ [ "request", "postcode" ], [ "request", "phone" ], [ "request", "address" ] ],
+		"system_utterance": "Yu Garden is a chinese restaurant in the east area.",
+		"turn_idx": 3,
+		"belief_state": [ { "slots": [ [ "slot", "postcode" ] ], "act": "request" }, { "slots": [ [ "slot", "phone" ] ], "act": "request" }, { "slots": [ [ "slot", "address" ] ], "act": "request" }, { "slots": [ [ "food", "chinese" ] ], "act": "inform" }, { "slots": [ [ "area", "east" ] ], "act": "inform" } ],
+		"user_utterance": "Could I get the address, phone number, and postcode of Yu Garden?",
+		"system_acts": []
+	},
+	{
+		"turn_label": [],
+		"system_utterance": "Phone is 01223 248882, address and postcode are 529 Newmarket Road Fen Ditton C.B 5, 8 P.A",
+		"turn_idx": 4,
+		"belief_state": [ { "slots": [ [ "food", "chinese" ] ], "act": "inform" }, { "slots": [ [ "area", "east" ] ], "act": "inform" } ],
+		"user_utterance": "Thank you. That is all the information I needed. Bye bye!",
+		"system_acts": []}
+]
+"""
+
+features = datasets.Features(
+    {
+        "dialogue_idx": datasets.Value("int32"),
+        "dialogue": [
+            {
+                "turn_label": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+                "system_utterance": datasets.Value("string"),
+                "turn_idx": datasets.Value("int32"),
+                "belief_state": [
+                    {
+                        "slots": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+                        "act": datasets.Value("string"),
+                    }
+                ],
+                "user_utterance": datasets.Value("string"),
+                "system_acts": datasets.Sequence(datasets.Sequence(datasets.Value("string"))),
+            }
+        ],
+    }
+)