Class 7 - Social robots Flashcards
Referential indeterminacy
situation in which different people vary in naming objects
Cross-situational learning
Mechanism for learning the meaning of words across multiple exposures, despite exposure-by-exposure uncertainty as to the word’s true meaning.
T/F: Cross-situational learning is considered a supervised learning model
False, it is considered unsupervised
Does cross-situational learning cope well with synonyms?
No
Does cross-situational learning cope well with inconsistent input?
No
What methods can we use to solve the scaling problem with inconsistent input (in the context of cross-situational learning)?
Scaffold model
Mutual exclusivity
Principle of contrast
Social mechanisms
Developmental robots
Robots used to test and validate theories about the role of interacting with the physical world on child language development.
Developmental robots can…
- be implemented with a theoretical model based on child behavior
- implement on the robot an AI theory about learning and see if they perform the same way as children in a learning task
N3C law
predicts learner will learn in all conditions
Mutual Exclusivity
states that children learn better in 2-competitor conditions
The 2 main conclusions of the paper by Twoney et al. are that…
- mutual exclusivity can account for learning
- explains how associative learning mechanism (matching a symbol / spelled word to its referent) can explain the “emergence” of mutual exclusivity
Referent selection in children
choosing the referent of a new word in an ambiguous environment
T/F: in robots, the learning effect can be inferred by looking at the robot’s activation function during a learning task
True
from lecture: “the decay in activation function reduces effect of learning when visual process takes longer.”
T/F: Language is grounded in interactions with physical
world
True
T/F: Embodied interactions may improve learning gain
True
In the context of effective tutoring, what does it mean that “interactions should be pedagogically sound”? (pick one)
A. ensuring that interactions provide qualitative input to optimize learning gain
B. children need to want to interact with the tutor for a prolonged period of time
C. children need to remain motivated to learn new skills
A. ensuring that interactions provide qualitative input to optimize learning gain
Multimodal interaction
using different modalities to communicate, i.e., gestures, pointing to objects, iconic gestures
Gesture-enriched vocabulary learning can…
- support vocabulary learning in adults & children (T/F)?
- work for different languages (T/F)?
- have a short-and long-term effect (T/F)?
True
True
True
Two important reasons why the robot should be seen as “peer” from the child’s point of view rather than a “teacher” during children-robot tutoring…
- the robot can make deliberate mistakes, allowing the children to correct it
- allows the robot to provide constructive negative feedback
What does “common ground” stand for in the context of child-robot interaction?
The robot and the child know things about each other and what to expect from each other
T/F: a common ground is established during the first encounter between child and robot and does not get updated during future interactions
False, it gets updated
Short-term common ground
During a child-robot interaction, going back to events that happened in the close past where both child and robot were present and refer to them.
Long-term common ground
Knowing things about each other’s past (child and robot)
T/F: Overall, personalization of a a child-robot interaction improves the learning experience
True
