Comprehending Words

Question 1

What is memory access like for word recognition?

Answer 1

Is it like a search where you look through your lexicon to find the right word? No, that is impossible because it would take too long, and we know word recognition is fast

In reality, word recognition is through a parallel activation representation/computation. As sounds are produced, our linguistic systems have phoneme and morpheme detectors that are activated as information is being said.

Rapid and happens in “real time”

Question 2

The Activation Model in Spoken Word Recognition

Answer 2

As you are hearing sounds (phonemes), the brain will activate multiple words that match the input for a brief amount of time

For example, we hear the phoneme /k/ and our brain triggers all the possible words that start with /k/ like cap, cat, cad. Then we get more information with /a/ phoneme and it keeps those words activated. Until we hear the last phoneme /t/, and now have only cat activated.

Question 3

What does the activation model predict?

Answer 3

Lexical competition among phonological neighbors
Parallel Activation of Homophones

Question 4

What are phonological neighborhoods?

Answer 4

Words that are only one phoneme away from each other
There are crowded phonological neighborhoods and sparse phonological neighborhoods

Question 5

What is a crowded phonological neighborhood?

Answer 5

Neighborhoods (a word) that contain a large amount of phonological neighbors
Like cat (cap, cad, hat, pat, mat, fat, kit, cut, etc)

They should be hard to identify in a crowd
(THINK: Where’s Waldo?)

Question 6

What is a sparse phonological neighborhood?

Answer 6

Neighborhoods (a word) that contain a small amount of phonological neighbors
Like stench (bench, wench, stent)

They should be easy to identify

Question 7

Neighborhood density effects

Answer 7

Experimental results demonstrating that it is more difficult and time-consuming to retrieve a word from memory if the word bears a strong phonological resemblance to many other words in the vocabulary than if it resembles only a few other words.

Question 8

Experiment examining Neighborhood and Frequency Effects in Word Recognition (Luce, 1986)

Answer 8

Procedure
Participants had to try to recognize (report) 800 consonant-vowel-consonant words, presented in static noise.
The words consisted of 200 words that were low spoken word frequency and from a dense neighborhood; 200 words that were high spoken word frequency and from a dense neighborhood; 200 words that were low spoken word frequency and from a sparse neighborhood; and 200 words that were high spoken word frequency and from a sparse neighborhood.

Findings:
It was the easiest to recognize high-frequency word in a sparse neighborhood (64%).
It was the hardest to recognize low low-frequency word in a dense neighborhood.

Question 9

Homophones

Answer 9

Two or more words that have separate, non-overlapping meanings but sound exactly the same (even though they may be spelled differently)

EX. Watch, verb vs noun meaning

Question 10

Homographs

Answer 10

Words that are spelled exactly the same but have separate, non-overlapping meanings (and may or may not sound the same)

Question 11

Cross-modal priming task

Answer 11

1. Hear sentence ending.
2. While hearing the sentence, you are seeing a word on a screen.
3. Judge if the word is real or not.

Question 12

Parallel Access of Meaning (Homophones) Study

Answer 12

Procedures: Cross-modal priming task
Hear a sentence ending with a homophone that either supports the noun meaning or verb meaning of the word watch.
While hearing the sentence, you are seeing a word on a screen. The word is either semantically related to the verb (look) or the noun (time) of the homophone watch or an unrelated word completely (dart).
The words are also presented at two different times; they are either presented immediately after the sentence ends or presented 250 msec after the sentence.
Judge if the word is real or not.

Findings:
They are faster to respond to both semantically related words compared to the one not semantically related when presented immediately after the sentence ends.

They are faster to respond to the semantically related word when the word is flashed 250 msec after the sentence ends.

Conclusion:
This supports the idea that people consider both meanings (parallel accessing for word recognition) before picking one meaning based on the context. There is parallel access when the semantically related words for both means are chosen right after the homophone is uttered, but when there is time (250 msec) to consider the context, then they are faster when looking at the word that is semantically related to that specific context

Suggests bottom-up activation followed by contextual selection
Suggests timing matters

Question 13

How is word recognition activation over time?

Answer 13

Cohort Theory of Spoken Word Recognition
TRACE Theory

Question 14

Cohort Theory of Spoken Word Recognition

Answer 14

Hearing the initial phonemes of a word will activate in parallel a cohort of words that share the initial onsets.
Further phonemes restrict the cohort of words.
Until the uniqueness point is reached, it is not possible to choose a single word that is being said.

Question 15

Cohort competitors

Answer 15

Words with overlapping onsets

Question 16

Uniqueness point

Answer 16

The point at which there is enough information in the incoming speech stream to allow the hearer to differentiate a single word candidate from its cohort competitors

Question 17

Study Supporting Cohort Theory

Answer 17

Procedures: Visual World Paradigm
Track eye movements of listeners as they hear instructions like: “Put the breaker above the square,” with a chess setup present. There were clip-art images of a beaker (target word), a beetle (cohort competitor), and a carriage (unrelated word).
So, trying to see what people look at while the sentence is being uttered.

Findings:
All the participants eventually get to the target word.
People are more likely to look at a cohort competitor than an unrelated word.
When the target word (beaker) is uttered, both the competitor and target are fixated around the same proportion. But by the offset of the target word then people diverge and more are proportionally looking at the target.

Question 18

What are the limitations of the Cohort theory?

Answer 18

Recognition of a word depends on hearing the first phoneme correctly so it is fragile

Question 19

What is an alternative theory to the cohort theory?

Answer 19

TRACE theory

Question 20

TRACE theory

Answer 20

The interactive activation model between levels
Works on continuous speech
Activates candidate words as speech unfolds, including middle words, rhymes
Thus allowing it to work even if we don’t hear the first phoneme well
Thus we have late activation of rhymes and initial activation of onset cohort

Question 21

Evidence for TRACE

Answer 21

Procedures:
Track eye movements of listeners as they hear instructions like: “Put the breaker above the square,” with a chess setup present. There were clip-art images of a beaker (target word), a beetle (cohort competitor), a speaker (rhyme), and a carriage (unrelated word).
So, trying to see what people look at while the sentence is being uttered.

Findings:
All the participants eventually get to the target word.

When the target word (beaker) is uttered, both the competitor and target are fixated around the same proportion, but there is also some proportion looking at the rhyme. The rhyme effect is greater than unrelated, later and smaller than the cohort competitor as predicted by TRACE

Question 22

TRACE computer model output

Answer 22

Proves TRACE because its computer model is highly consistent to the behavior of human eye movements with 93% of variance accounted for

Question 23

Semantic priming

Answer 23

The phenomenon by which hearing or reading a word partially activates other words that are related in meaning to that word, making the related words easier to recognize in subsequent encounters

Question 24

Lexical decision task

Answer 24

An experimental task in which participants read strings of letters on a screen that might either be actual words or nonsense words. Subjects press one button if they think they’ve seen a real word, or a different button to signal that the letters formed a nonsense word. Response times for real words are taken as a general measure of the ease of recognizing those words under specific experimental conditions

Question 25

Facilitation

Answer 25

Processes that make it easier for word recognition to be completed

Question 26

Inhibition

Answer 26

Processes that result in word recognition becoming more difficult

Question 27

Mediated semantic priming

Answer 27

The process by which a prime word speeds up responses to a target word not because of a direct connection but due to an indirect connection via some other intervening word

Question 28

Excitatory connections

Answer 28

Connections along which activation is passed from one unit to another, so that the more active a unit becomes, the more it increases the activation of a unit it is linked to.

Question 29

Inhibitory connections

Answer 29

Connections that lower the activation of connected units, so that the more active a unit becomes, the more it suppresses the activation of a unit it is linked to.

Question 30

Incremental language processing

Answer 30

The processing of language in such a way that hearers begin to generate hypotheses about the meaning of the incoming speech on the basis of partial acoustic information, refining and revising these hypotheses on the fly rather than waiting until there is enough information in the speech stream for the hearer to be certain about what the speaker meant