MOVIE 4 TRANSCRIPT

Example shows OFAI's dialogue-driven training of the Adaptive Mind Rascalla, plus DFKI's anaphoric reference, plus OFAI's news mechanisms based on
a) monitoring and of RRS-Feeds and
b) communication to other Rascalli;
a) and b) build upon information provided by ARC/RSA's Agent Modelling Server

download/view


transcript:
U ... User
R ... Rascalla/a

U: "Who is Rihanna?"
(processing steps:
1. U is processed by OFAI's T-IP4Adaptive;
2. OFAI's Adaptive Mind component selects T-ChatBot as actuator tool for answering the user question;
3. the multimodal answer is generated by OFAI's T-MMG, the resulting BML is sent to the Nebula Client for animation)
R (resulting from the bot): "I do not recognize the name …"
(user colds the agent)
R: "I tried my best."
U: "Who is Rihanna?"
R (multimodal):
a) multimodal speech: "These links will probably tell you more."
b) presentation of links retrieved by making use of OFAI's open-domain QA
U: "Tell me more"
(This user utterance signals the Adaptive Mind that the previous answers was good, and that the user wants to know more about Rihanna. Having a memory about which tools (actions) the agent has already tried to answer the current question, the Adaptive Mind selects a new tool, in the current case the ChatBot, to answer "Who is Rihanna?")
R: "No one that I have talked to …"
U: "Try again"
(the user signals that she is not satisfied with this answer, and the agent should try another, not yet used, tool; having already applied the open-domain QA (OFAI) and the ChatBot (OFAI), now the mind selects DFKI's gossip answer tool)
R (multimodal):
a) multimodal speech: "Here, have a look …"
b) presentation of artist information via DFKI's Visual Browser Web application
U: "Do you have any news about her?"
(the anaphoric reference is resolved to Rihanna by DFKI's anaphora resolution component, together with the analysis from OFAI's T-IP; enough information is provided to the Adaptive Mind so that it triggers the text-relevance tool to find a) news feeds providing information about Rihanna and b) other Rascalli that have news about Rihanna; both fail for reasons being explained to user by the Rascalla)
R: "I am sorry, I do not have a feed URL and the agents I asked could not help me. …"
(long tail of remaining movie that needs to be cut off)

Movie 2 - example of the clarification dialogue and following presentation of the correct artist information

U: Who is Madonnna?
R: "Sorry, I don't know that. Did you perhaps mean Madonna instead of Madonnna?"
U: "yes"
R (multimodal presentation):
a) multimodal utterance: "Here, have a look at the personal profile of Madonna."
b) presentation of artist information via DFKI's Visual Browser Web application

Movie 3 - example of the integration of DFKI's gossip dialogue system with OFAI's Adaptive Mind Music Companion)

U: "Who is Tina Turner?"
(processing steps:
1. User utterance is processed by OFAI's T-IP4Adaptive incorporating DFKI's question analysis component for gossip dialogue;
2. OFAI's Adaptive Mind component selects DFKI's answer generation for gossip dialogue as actuator tool;
3. the output of DFKI's answer generation tool is passed through OFAI's T-MMG which provides BML input to the Nebula Client)
R (multimodal answer):
a) multimodal speech: "Here, have a look at the personal profile of Tina Turner"
b) presentation of link to the artist page in DFKI's Visual Browser
U: "Who is her husband?"
(processing steps:
1. U is processed by OFAI's T-IP4Adaptive incorporating DFKI's question analysis component for gossip dialogue which provides anaphoric reference between her and Tina Turner;
2. OFAI's Adaptive Mind component selects DFKI's answer generation for gossip dialogue as actuator tool;
3. the output of DFKI's answer generation tool is passed through OFAI's T-MMG which provides BML input to the Nebula Client)
R (multimodal speech): "To my knowledge, …"
(User praises the Rascalla)

Movie 4 - example for combination of try again and clarification dialogue

U: Who is Eminen?
R: "No one that I have talked to …"
(this is a reply from the ChatBot)
U: "Try again"
(the user indicates that she is not satisfied with the answer and that the Rascalla should try another tool; this triggers the Adaptive Mind to use a tool other than ChatBot, in our case the DFKI answer generation tool is selected which realizes that Eminen might be a typo and therefore the Rascalla starts a clarification dialogue)
R: "Sorry, I don't know that. Did you perhaps mean Eminem instead of Eminen?"
U: "Yes"

RASCALLI is supported by the European Commission Cognitive Systems Programme (IST-27596-2004).

 
 
Internal links: