cs 182 sections 103 - 104 slides created by eva mok ([email protected])[email protected]...
Post on 19-Dec-2015
219 views
TRANSCRIPT
CS 182Sections 103 - 104
slides created by Eva Mok ([email protected])
modified by JGM
March 22, 2006
The Last Stretch
Cognition and Language
Computation
Structured Connectionism
Computational Neurobiology
Biology
MidtermQuiz Finals
Neural Development
Triangle Nodes
Neural Net & Learning
Spatial Relation
Motor Control Metaphor
SHRUTI
Grammar
abst
ract
ion
Regier Model
Bailey Model
Narayanan Model
Chang Model
Visual System
Psycholinguistics Experiments
Quiz
• How does Bailey use multiple levels of representation to learn different senses of verbs?
• What’s the difference between aspect and tense?
• What’s X-schema embedding? Give an example where embedding is necessary.
• How can Reichenbach’s system be used with X-schemas? When does this system break down?
1. What does the prefix affix tree look like for the following sentences:
– eat them here or there
– eat them anywhere
2. What does the affix tree look like after the best prefix merge?
Quiz
• How does Bailey use multiple levels of representation to learn different senses of verbs?
• What’s the difference between aspect and tense?
• What’s X-schema embedding? Give an example where embedding is necessary.
• How can Reichenbach’s system be used with X-schemas? When does this system break down?
1. What does the prefix affix tree look like for the following sentences:
– eat them here or there
– eat them anywhere
2. What does the affix tree look like after the best prefix merge?
Bailey’s VerbLearn Model
• 3 Levels of representation
1. cognitive: words, concepts
2. computational: f-structs, x-schemas
3. connectionist: structured models, learning rules
• Input: labeled hand motions (f-structs)
• learning:
1. the correct number of senses for each verb
2. the relevant features in each sense, and
3. the probability distributions on each included feature
• execution: perform a hand motion based on a label
Computational Details
• complexity of model + ability to explain data
• maximum a posteriori (MAP) hypothesis
)|( argmax DmPm
rule Bayes'by )()|( argmax mPmDPm
how likely is the data given this model?
penalize complex models – those with too many word senses
wants the best model given data
schema elbow jnt posture accel
slide extend palm 6
schema elbow jnt posture accel
slide extend palm 8
schema elbow jnt posture accel
slide 0.9 extend 0.9 palm 0.9 [6]
data #1
data #2
data #3
data #4
schema elbow jnt posture accel
depress 0.9
fixed 0.9 index 0.9 [2]
schema elbow jnt posture accel
slide 0.9 extend 0.9 palm 0.9 [6 - 8]
schema elbow jnt posture
slide 0.9 extend 0.9 palm 0.7
grasp 0.3
schema elbow jnt posture accel
depress fixed index 2
schema elbow jnt posture accel
slide extend grasp 2
Limitations of Bailey’s model
an instance of recruitment learning (1-shot)
embodied (motor control schemas)
learns words and carries out action
the label contains just the verb
assumes that the labels are mostly correct
no grammar
Quiz
• How does Bailey use multiple levels of representation to learn different senses of verbs?
• What’s the difference between aspect and tense?
• What’s X-schema embedding? Give an example where embedding is necessary.
• How can Reichenbach’s system be used with X-schemas? When does this system break down?
1. What does the prefix affix tree look like for the following sentences:
– eat them here or there
– eat them anywhere
2. What does the affix tree look like after the best prefix merge?
Aspect
• Aspect is different from tense in that it deals with the temporal structure of events
• Viewpoints
– looking at the same event at different granularity
– He was walking / He has walked / He walks
• Phases of Events
– zooming in at a level and focusing on a stage in an event
– He is about to walk / He finished walking
• Inherent Aspect
– perfective / imperfective (telic / atelic)
– He is walking / He is tapping his finger
Some linguistics theories on Inherent Aspect
• Zeno Vendler (1957)’s distinction on state, activity, accomplishment, achievement
stative
Verbal predicates
dynamic
atelic atelic telic
protracted instantaneousknowresemble
runswim
write a letterrun a mile
jumprecognizestate activity
accomplishment achievement
FYI:telic = boundedatelic = unboundedpunctual = instantaneous
Controller X-Schema• The controller x-schema is meant to capture the generic structure of events.
• Aspect therefore marks (or profiles) certain states or transitions.
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
The car is on the verge of falling into the ditch.
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
FALL
He stumbled on the uneven road.
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
WALK
She cancelled her trip to Paris.
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
TRAVEL
Quiz
• How does Bailey use multiple levels of representation to learn different senses of verbs?
• What’s the difference between aspect and tense?
• What’s X-schema embedding? Give an example where embedding is necessary.
• How can Reichenbach’s system be used with X-schemas? When does this system break down?
1. What does the prefix affix tree look like for the following sentences:
– eat them here or there
– eat them anywhere
2. What does the affix tree look like after the best prefix merge?
X-Schema Embedding
You can ‘blow up’ any state or transition into a lower level x-schema, allowing embedding
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
readystart
ongoingfinish
done
He is almost done talking.
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
readystart
ongoingfinish
done
They are getting ready to continue their journey across the desert.
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
readystart
ongoingfinish
done
She smokes. (habitual reading)
readystart
ongoingfinish
done
iterateabortcancelled
interrupt resume
suspended
readystart
ongoingfinish
done
Quiz
• How does Bailey use multiple levels of representation to learn different senses of verbs?
• What’s the difference between aspect and tense?
• What’s X-schema embedding? Give an example where embedding is necessary.
• How can Reichenbach’s system be used with X-schemas? When does this system break down?
1. What does the prefix affix tree look like for the following sentences:
– eat them here or there
– eat them anywhere
2. What does the affix tree look like after the best prefix merge?
Mapping down to the time line
• we can use Reichenbach’s system to map the controller X-schema down to a time line and get tenses
S R E
readystart
ongoingfinish
done
Speech Time (S)
Reference Time (R)
Event Time (E)
He would have talked...
• … by the time the bell rang.
• … if you had not shushed him.
• okay this is getting complicated
• interaction with modals like would
• could be a strictly past reading (case 1)
• possibly a counterfactual (case 2)
Not much of a stretch…
to get to metaphorical sentences like this:
The US Economy is on the verge of falling back into recession after moving forward on an anaemic recovery.
• You would just need to map from this physical domain (source domain) to the say, economics (target domain)
• the Event Structure Metaphor is exactly the general mapping from motion to changes, locations to states
• and then you need some domain specific mappings
• more on that in lecture…
Quiz
• How does Bailey use multiple levels of representation to learn different senses of verbs?
• What’s the difference between aspect and tense?
• What’s X-schema embedding? Give an example where embedding is necessary.
• How can Reichenbach’s system be used with X-schemas? When does this system break down?
1. What does the prefix affix tree look like for the following sentences:
– eat them here or there
– eat them anywhere
2. What does the affix tree look like after the best prefix merge?
Affix Trees
• Data structures that help you figure out what merges are possible
• Each node in the tree represents a symbol, either terminal or non-terminal (we call that the “affix” in the code)
• Prefix Tree
• Suffix Tree
Prefix Tree
r1: S eat them here or therer2: S eat them anywhere
eat
them
here
or
there
r1
r1
r1
r1
r1
r2
anywhere
r2
r2
Sr1 r2
Prefix Merge
r3: S eat them X1r4: X1 here or therer5: X1 anywhere
eat
them
r3
r3
S
r3
here
or
there
r4
r4
r4
anywhere r5
X1
X1r3
r4 r5