Learning Words from Multiple Situations

Question 1

What is a global learner?

Answer 1

A global learner stores everything, all possible references for each instance of the word (encodes the situation in which they heard the word)

Then, he recalls the situation when he hears the word again and then aggregates. Whatever is present in both situations has to be what the word represents.

Question 2

What is a local learner?

Answer 2

A local learner encodes what he thought the word meant within a given context.

Then recalls that meaning the next time he hears the word (retrieve what they remember the word meant), And then revises their definition of the word if necessary based on the current situation.

Question 3

Yu & Smith Study Evidence for Global Learning?

Answer 3

Procedures:
Asked adults to learn words from referentially ambiguous situations

There are two conditions with differ with the referential ambiguity objects being presented. One condition had 2 objects with 2 words, and the other condition had 3 objects with 3 words.

There was a learning phase in which the subjects were presented with two objects and heard two words, then were presented with two new objects and heard two new words. After several trials, a word and an object from the past trials reappear. This process is repeated with the second condition with three objects.

After comes the test phase, in which the subjects are presented with four objects from the learning phase and told to “Click on the dax.”

Prediction:
In the 3x3 learning phase, it co-occurred with a lot more noise (more possible references)
So if people are global learners harder to learn 3x3 than 2x2 with more referent ambiguous reference

Findings:
Consistent with a global learner who recalls what co-occurred with DAX during the learning phase.
In the 2x2 condition, the proportion correct is the most accurate.
All the conditions are above chance.

Question 4

Yu & Smith Follow-Up Study Evidence for Local Learner?

Answer 4

Aim: Originally thought to be a global learner model, but the local learner model makes the same prediction

Procedures:
Subjects were shown two objects and two words.
Based on the local learner model, the learner may choose the correct object for dax and then be presented with the object again with dax. The situation further strengthens the association of the object with dax.

But if in the first situation with the objects with the word dax, the subject guessed incorrectly. When they recounter the word dax, they would retrieve their first guess of the can, see that it is not present, so they make a random guess but have a 50% chance. Yet when done in the 3x3, the subject has a lower probability of guessing the correct object since there are now three objects instead of two to try to guess correctly.

Thus, the 2x2 condition is more likely to stumble correct meaning here, and the 3x3 condition is less likely to stumble correct meaning here

Question 5

What happens when the learner is faced with a series of natural word-learning situations like those encountered by children in the home?

Answer 5

We use a human simulation paradigm study

Question 6

Medina, Snedeker, Trueswell & Gleitman Study

Answer 6

Procedures:
The study uses the human stimulation paradigm.
Adults watch muted videos of parents talking to their children in the home.
Each vignette is a present saying a particular common noun.
At that moment, the parent uttered the mystery noun; adults instead hear a nonsense word (vash)
After watching the video, they are asked to guess what vash means.

Implications:
The different vignettes allow for learning across situations from natural contexts.
To see whether there is a global or local learner

Findings:
People only remember their previous guess, not the previous situation.

So, the subject’s response on the previous vignette for that word was correct, which led to a 0.45 proportion correct of the current vignette.
If the response to the previous vignette for that word was incorrect, the proportion of accuracy on the current vignette was just chance.

If a separate group that didn’t receive any explicit feedback (didn’t get cross-situation comparison), they performed below chance.

Question 7

Woodard, Gleitman, and Trueswell - Children as local learners

Answer 7

Procedures:
Told 2 and 3-year-old children going to an animal Safari, and be able to see animals. There are two conditions, in which one lets children be correct, and the other condition lets children be incorrect.

In the let children be correct condition, the children see 2 novel animals that they have never seen, and are told I see a mipen, point to the mipen. Then they point to the animal. Then, two intervening trials involving known animals are presented. After they see, their previous guess for mipen and a new novel animal.
The prediction should be that the child should pick the same critter again.

In the let children be incorrect condition, the children see 2 novel animals that they have never seen, and are told I see a mipen, point to the mipen. Then they point to the animal. Then, two intervening trials involving known animals are presented. After they hear again “I see another mipen, point to the mipen, yet it has the animal they did not pick in the last trial, and a new novel animal.
The prediction is that if the child doesn’t remember what else co-occurred with mipen, she should now be at chance if a local learner.
If a global learner, they should have stored the previous context.

Findings:
Children are very local learners
The accuracy is higher than chance in the let the children be correct trial.
The accuracy in the incorrect trial is below chance, which shows that children can’t recall the previous content (situation) of the mipen.

Question 8

Why minimal cross-situational comparison (global learner)?

Answer 8

Listeners/learners believe an utterance has just one intended referent
If people didn’t commit in the moment and instead tracked everything, learning becomes diluted.

Question 9

What is the dilution effect?

Answer 9

If we were to track everything, to try to learn what the word means from multiple situations, we have to keep track of everything, which leads to the probability mass being highly distributed of what the word (dax) means

Question 10

How do we avoid the dilution effect?

Answer 10

By being a local learner, in that we just remember what we guessed the word meant

Question 11

What do the findings of the different studies suggest>

Answer 11

Adults are local learners
Children are very local learners

Question 12

So overall

Answer 12

The perceptual world offers way too much information for word learning

Language learner reduces referential alternatives by focusing on a single hypothesis, which is an effective way of avoiding dilution