22 - Word Meaning 2

Question 1

What are limitations with learning referentially?

Answer 1

We cannot learn abstract concepts (like ‘mind’ ‘thoughts’) and words can mean different things

Question 2

What is Chompsky’s (1980) “Poverty of the Stimulus”?

Answer 2

We acquire the majority of words we learn indirectly

We learn words before we know EXACTLY what they mean (we vaguely know)

There must be something innate in word learning

We take more meaning from a sentence than what is contained in a sentence

12-year-olds learn 10-15 words per day, <1 of these through direct conversation

(most through reading)

This would require reading approx. 20 paragraphs per day, but the average amount for this age is 5

This is the problem of induction

Question 3

We not only understand the meaning of words, we understand the ________ ________ _____

Answer 3

Relationship between them

E.g. dog refers to the object dog, but also other things

Question 4

What are the 4 concept relations that all contribute to our understanding of a word?

Answer 4

Word-concept relation

Concept-concept relation

Concept-perception relation

Word-word relation

Question 5

Are the 4 concept relations independent of one another or are all relevant?

Answer 5

All are relevant to understanding a word

Question 6

What is a word-concept relation?

Answer 6

What concept does the word relate to

E.g. dog = animal

Question 7

What is a concept-concept relation

Answer 7

What are the other features of the concept

E.g. dog = barks and has a tail

Question 8

What is a concept-perception relation?

Answer 8

What are our perceptions of a dog and what makes up our schema of a dog

E.g. dog = we know what dog’s look like and they sniff other dog’s bums

Question 9

What is a word-word relation?

Answer 9

Words that are associated with the word

E.g. dog is associated with bone, opposite to cat

Question 10

When do we begin to be able to understand words at a more conceptual level?

Answer 10

Around 7 years of age

Question 11

Children under 7 show “syntagmatic responding” with word associations. What does this mean?

Answer 11

Completing a sentence

E.g. if you said “cold” they might say “outside”

Question 12

Children over the age of 7 show “paradigmatic responding” with word association. What does this mean?

Answer 12

Creating an association with a relating word

E.g. “Cold” is associated with “hot”

Question 13

What are the two views regarding why children begin to show paradigmatic responding at age 7?

Answer 13

Maturation effect: as children get older their brains develop to make these shifts

First to second-order associative learning (statistical learning): we track the distribution of the co-occurrence of words. Based on this idea children are learning to read and therefore begin to connect these words together.

This might explain the poverty of the stimulus and induction

Question 14

Adults who are learning a second language go from syntagmatic to paradigmatic responses the better they get at learning a language. What view explains why this happens better?

Answer 14

First to second-order associative learning

Question 15

What is first to second-order associative learning?

Answer 15

Word meaning can be derived from the statistical pattern derived from the context of which it occurs.

• Quotes:*
• “You shall know a word by the company it keeps”*
• • Firth (1957)*
• “If words A and B have almost identical environments, they have similar*
• Meaning”*
• • Harris(1954)*
• “Verbal meaning = activation of the relations between words”*
• • Landauer (2007)*

Question 16

Describe the “distributional theories of word meaning”

Answer 16

We use statistical learning to track how words co-occur together

Distributed linguistic representations: the meaning of a word is dependent on the distribution with which it occurs with other words

Frequency of co-occurrences is not enough, co-occurrence distributions determine word meaning as similarity abstracted over context

This has been tested through computational models that tracks word co-occurrences on a large scale

Question 17

How have distributional theories of word meaning been tested?

Answer 17

Computational models call “latent semantic analysis” (LSA)

Question 18

What is latent semantic analysis (LSA)?

Whats a similar model?

Answer 18

Tracks the co-occurrence of words and nothing else

Compress data: Lose information but learn about relations

Hyperspace Analogue to Language (HAL) (another similar model)

Looks at large amounts of text to determine how often a word appears in a text

The analysis is devoid of all meaning, just frequency

Question 19

Do computational models of distributional theories that match words together have good validity?

Answer 19

Yes, they are quite good at representing our behaviour as humans

Question 20

When latent semantic analysis (LSA) was applied to intelligence tests, what were the results?

What theory did this provide evidence for?

Answer 20

When it was used in the vocabulary section of general knowledge tests it was the single best predictor of general intelligence

LSA correct on 64% of items

• Matches average applicant to US college
• • Mistakes correlate with those of the participants (r = .44)*

Evidence for the distribution theory

Question 21

What is “priming”?

Answer 21

When your brain is preparing associative relationships with words unconsciously

Makes us faster at responding to the world

Question 22

What is a “priming” test?

Answer 22

Determined by a lexical decision task

Look at two words momentarily and decide whether the second word is a real English word

The speed from priming is determined by the associative relationship

Question 23

How does distribution theory explain priming?

Answer 23

People were faster at decoding the second word in a lexical decision task if the word is primed with an associative English word

Question 24

Are word associations produced faster if they are semantic (DOG-WOLF), associative (DOG-BONE) or “both (UNCLE-AUNT)?

Answer 24

Associative (machine learning is the same)

Question 25

What is the “symbol grounding problem”

Answer 25

There would be no connection between written words without a mind to interpret them and make the connection.

• Word meaning is made in the brain and not outside in the world*
• How can humans therefore understand word meaning the same way as computational models?*

Question 26

The “symbol grounding problem” led to which view of word association?

Answer 26

The embodied view

Question 27

What is the “embodied view” of language meaning

Answer 27

Meaning is grounded in bodily states

Our understanding of language is ground in our body and what it can do

Meaning reflects affordances: relations between objects and bodily states

There is a direct connection between the symbol we read and internal perceptual representations/ simulations

This means the same neural systems engaged in perception and action are automatically activated during semantic representation

Question 28

According to the “embodied view” of language meaning, all the symbols we know as words are connected to our _____ and _____.

Answer 28

Perceptions and actions

The same neural connections occur when we think of something or actually experience it

Question 29

What are the 3 steps to process a sentence according the “embodied view”

Answer 29

1. Connect the words (symbols) to internal perception
2. Derive affordances
3. Combine all of this information into a mental model

Question 30

What is the major argument in favour of the “embodied view”?

Answer 30

The use of affordances

• i.e. All the possible actions a human or object can possible take*
• E.g. humans can sit or stand, run or jump but goldfish can’t do these things*
• Therefore these are affordances that are afforded to us but not for goldfish*

Question 31

Embodied Approach Example:

Mike was freezing while walking up State Street into a brisk wind. He knew that he had to get his face covered pretty soon or he would get frostbite. Unfortunately, he didn’t have enough money to buy a scarf.

~ Being clever, he walked into a store and bought a newspaper to cover his face

What is this an example of in the embodied approach?

Answer 31

Afforded but not related

Realistic but not valid

Question 32

Embodied Approach Example 2:

Mike was freezing while walking up State Street into a brisk wind. He knew that he had to get his face covered pretty soon or he would get frostbite. Unfortunately, he didn’t have enough money to buy a scarf.

~ Being clever, he walked into a store and bought a matchbox to cover his face.

What is this an example of in the embodied approach?

Answer 32

Non-afforded

Not realistic

Question 33

Embodied Approach Example 3:

Mike was freezing while walking up State Street into a brisk wind. He knew that he had to get his face covered pretty soon or he would get frostbite. Unfortunately, he didn’t have enough money to buy a scarf.

~ Being clever, he walked into a store and bought a ski-mask to cover his face

What is this an example of in the embodied approach?

Answer 33

Both afforded and related

Realistic and valid

- would do a great job of covering face and is related to the first sentence

Question 34

Studies have shown that we are equally able to visualise afforded, non-afforded and related sentences

• true or false

Answer 34

False, we are much better at imagining afforded and related sentences than non afforded

Question 35

Can LSA models account for affordances?

Answer 35

No, humans can do but computer models can not.

Question 36

Neuroscience and MRI scans support which view of language meaning?

How do they show this?

Answer 36

Embodied view

They are able to show the activation in our brains and the neural region they are associated with

• E.g. if you say the word “kick”, the related sensory-motor system for moving the leg is activated*
• “Wipe” is associated with the activation of the neural region associated with the movement of the participants preferred hand*
• This shows that our understanding of the meaning of language is embodied in some degree and this occurs within us, rather than out there in the world*

Question 37

How did analysis of patients with motor-neuron disease support the embodied view of language and meaning?

Answer 37

Patients with motor-neuron disease have more impaired verb comprehension and production that healthy controls and those with neural diseases in isolation.

Question 38

Why are abstract words a problem for the embodied view?

Answer 38

We don’t embody abstract words that cannot be perceived or experienced.

Question 39

How do defenders of the embodied view argue against abstract words being a problem?

Answer 39

They suggest that we embody them through metaphors

• (e.g. love = loveheart, freedom = bird flying)*
• This can’t explain differences though, doesn’t quite adequately represent it*
• What about jargon that doesn’t have visual representations?*

Question 40

What is the most accurate meaning of language model we have currently?

Answer 40

the dual process view

a combination of the computational model and the embodied model that gives us the best answer to how we derive meaning from language.

Question 41

Give an example of the “dual-process view”

Answer 41

An object activates a perceptual representation

e. g. when reading
- Activates distributional learning through co-occurrences (computational view)
- Activation of a perceptual representation is optional and might only happen when we are engaged (embodied view)
- Based on when required

Question 42

How are abstract words explained by the dual-process view?

Answer 42

Learned by tracking the co-occurrences (computational view)

Relationship to emotional states to help us ground meaning (embodied view)

Supported by the fact that the first abstract words children use are “good, bad, happy, sad” - emotional words

Question 43