Update README.md

README.md

@@ -8,23 +8,58 @@ pipeline_tag: text-generation
 This model requires instructions. Following is an example input sequence:
 
 ```
-Given is the following task-oriented dialog between a human user (<user>)
-and a virtual agent (<system>). Previously, this conversation went wrong 
-because the virtual agent made a statement that was contextually incorrect.
-The human user reacted accordingly (No, I don't think so. I asked for assistance
-with my legal dispute.). Generate the user's intent (<intent>), extract the
-slot values (<slots>) and generate the next system utterance by considering 
-the user's emotion (confusion). <dialog> <user> Hey there! How's it going? 
-<system> Hello! How can I assist you today? <user> I need assistance with my 
-legal dispute. Can you help me? <intent>
+You are a virtual agent specializing in postal services, insurance and reception. Your job is to guide customers through the process of parcel shipping,
+answer their questions about insurance or register them, open the turnstile and tell them where to find their meeting room.  To do this, you need to
+understand the customers' intentions and the information they provide in their uttrances in order to answer them in a helpful and friendly manner. 
+
+###Instruction
+Consider the following conversation between you and a customer. Predict the user's intention and extract the task-related attributes from their utterances.
+Generate your next answer, also considering the knowledge below. Return the results line by line. Here is an example:
+
+User Intention:
+    Parcel Choice
+Attributes:
+    Weight: 10kg
+    Destination: London, UK
+Virtual Agent:
+    If your item weighs only 10kg, I recommend to use our medium-sized box.
+
+For user intention, the following values are possible: Greeting,Parcel Choice, Recharge Phone, Building Access, Question Answering.
+For Attributes, the following values are possible: Outcome Operation, Bill Form Payment Procedure, Import Payment, Destination, Type of Bills, Host Name,
+Confirmation to Open the Turnstile, Delivery Option, Ticket Number, Verification Call, Weight, Phone Number, Meeting Date and Time, Bill Form Name, Shipping
+Box Description, Host Email, Shipping Procedure, Meeting Room Identifier, Guest Name, Confirmation to Open Turnstile, Phone Provider, Package Required,
+Alternative Host Email, Bill Form Description, Question, Type of Service, Alternative Host Name, Shipping Box Name, Shipping Time, Evidence.
+
+###Knowledge
+[knowledge document if available]
+
+###Conversation
+[dialogue history]
+
+The user waits for a response from the virtual agent.
+
+###Response
+
+User Intention:
 ```
 
-This is the expected output sequence:
+Please replace [knowledge document if available] with the knowledge document or an empty string. Please replace [dialogue history] with the dialogue context, e.g.:
 
 ```
-<intent> question_answering <slots> <question> I need assistance with my legal 
-dispute. Can you help me? <system> I apologize for the misunderstanding. How 
-can I assist you with your legal dispute?
+Customer: Hi there! 
+Virtual Agent: Hello! How can I assist you today? 
+Customer: I just adopted a cat and I'm interested in getting insurance coverage for accidents and illnesses. Which document should I refer to for information on this? 
+```
+
+This is an example for the expected output:
+
 ```
+###Response
 
-(The linebreaks are not necessary and just added to make the sequences more readable.)
+User Intention:
+    Question_answering
+Attributes:
+    Question:  I just adopted a cat and I'm interested in getting insurance coverage for accidents and illnesses. Which document should I refer to for information on this
+Virtual Agent:
+    You might want to check document_0, which outlines our coverage and assistance services in case of accidents or illnesses suffered by the Animal."
+```