Robot Interaction
Lecture 8
, Lecture & Literature - Summary
Understanding Language
Natural Language Understanding & Generation
3 main phenomena (all three are interrelated):
1. Identity
Detecting the things in the world that surrounds us.
2. Reference
How you can talk about these things (how can you make reference to these objects)
3. Perspective
The way we make reference is a function of the perspective that we have.
Problems in natural language
understanding/processing:
Ambiguity (dubbelzinnigheid)
o Arises when we refer to a single thing from many different perspectives.
o This is a problem for machines.
Variation
o We can make reference to a lot of things, the way we make references has a lot of
implications, there are many ways to refer to things, things can be expressed by
many words and sentences
Vagueness
o If there is a world that we observe, we never fully describe that world in language.
o We assume there is a common ground with a certain agent, and therefore we make
partial reference to another agent.
Since 1 simple sentence can have more than a quadrillion different meanings (since every single word
itself can have many different meanings, but also words linked to each other can have other
meanings) it is very hard for machines to understand such a meaning. Since humans have common
ground with other humans and based on that they can derive an abstract meaning of a sentence.
Machines don’t have such abilities.
There are many ways of learning through
conversation:
Robots need to learn from humans to establish sufficient common ground in language processing
(even though humans make mistakes as well).
Experience grounding
o Reinforcement learning
Doing something and copying that action
Symbolic relations
o Semantic web
Symbolic stories
o The news
Sensory observation feedback
o Looking around and inferencing things based on observations
Symbolic properties
o Talk
Lecture 8
, Lecture & Literature - Summary
Understanding Language
Natural Language Understanding & Generation
3 main phenomena (all three are interrelated):
1. Identity
Detecting the things in the world that surrounds us.
2. Reference
How you can talk about these things (how can you make reference to these objects)
3. Perspective
The way we make reference is a function of the perspective that we have.
Problems in natural language
understanding/processing:
Ambiguity (dubbelzinnigheid)
o Arises when we refer to a single thing from many different perspectives.
o This is a problem for machines.
Variation
o We can make reference to a lot of things, the way we make references has a lot of
implications, there are many ways to refer to things, things can be expressed by
many words and sentences
Vagueness
o If there is a world that we observe, we never fully describe that world in language.
o We assume there is a common ground with a certain agent, and therefore we make
partial reference to another agent.
Since 1 simple sentence can have more than a quadrillion different meanings (since every single word
itself can have many different meanings, but also words linked to each other can have other
meanings) it is very hard for machines to understand such a meaning. Since humans have common
ground with other humans and based on that they can derive an abstract meaning of a sentence.
Machines don’t have such abilities.
There are many ways of learning through
conversation:
Robots need to learn from humans to establish sufficient common ground in language processing
(even though humans make mistakes as well).
Experience grounding
o Reinforcement learning
Doing something and copying that action
Symbolic relations
o Semantic web
Symbolic stories
o The news
Sensory observation feedback
o Looking around and inferencing things based on observations
Symbolic properties
o Talk
