L&C, W3

Question 1

What is visual word recognition

Answer 1

• First stage in reading things > see letters on page + combine them to reach a meaning of the word
• A small set of symbols in combination makes up an infinite set of words. E.g. How many words can you make out of the 4 letters “C” “A” “T” “S”?
• What representations are used to access the mental lexicon? > mental lexicon = mental dictionary that contains information regarding a word’s meaning, pronunciation, syntactic characteristics (depending on age/language, usually have 60,000 to 70,000 words in mental lexicon)
- What are the units, calculated from the visual input, that are used to address the mental dictionary? Is it single letters? Grapheme (clusters)? Syllables? Morphemes? > essentially, when you see the word, what do you use from it (Visual input) to find the word in the mental lexicon

Question 2

Q1. Are graphemes used to access visual words?

Answer 2

• The name grapheme is given to the letter or combination of letters that represents a phoneme. For example, the word ‘ghost’ contains five letters and four graphemes (‘gh,’ ‘o,’ ‘s,’ and ‘t’), representing four phonemes.
• Graphemes are letters and letter groups that correspond to one sound (phoneme). Hence, they act as a ‘functional bridge’ between phonology (sound of word) and orthography (conventional spelling of word).
• Bread has 4 graphemes: B R EA D = 4
So do you use EACH letter (5) to go to your mental lexicon and decide it is a word OR do you use the graphemes (4) to decide what word it is

Question 3

Task detection: was the target letter present?

Answer 3

• Ppts were shown target letter A, ppts had to say very quickly when presented with a word whether there was an A in the word
• Ppt sees target letter (700ms), brief pause (1000 ms) and then the word is presented for only 33 ms (almost below conscious perception) > then mask + response
2 conditions: Condition 1: had the word BOARD, where OA is 1 grapheme. Condition 2: had the word BRASH where A is it on it’s own

Question 4

Q1. Possible outcomes + results

Answer 4

• If graphemes ARE perceptual units: Finding A should take longer in the condition 1 aka BROAD than in condition 2 aka BRASH > this is because the OA will need to be split up within BROAD
• If there are no perceptual grapheme units, the time taken to find A in BROAD and BRASH should be equal > this is because there is not a grapheme to split up
Are grapheme access units to the lexicon?
• The result shows that multi-letter graphemes aka OA in BOARD took longer than the single target letter A in BRASH to be detected
• Graphemes are processed as perceptual reading units > to get to the A in BOARD you need to break apart OA

Question 5

Q2. Are morphemes access units in reading?

Answer 5

• Morphemes are the smallest meaningful unit of a language
○ can be a word itself (e.g. deck) or part of a word (de-brief)
○ morpheme – roots and affixes (prefixes & suffixes)
○ “unreal” has two morphemes “un” (prefix -word/letter placed before another) and “real” (root - the main meaning of the word)
- “teaspoon” has two morphemes “tea” and “spoon” > each has individual meaning, no prefix + root > Do we read “unreal” / “teaspoon” as a single word, or through its parts “un”+”real” / “tea” + “spoon”?
• Pseudo-affix words are words which look like an affix but they are not actually one
• “swing”? Is this “sw” + “ing” > running + carrying are an affix but swing is not because sw on it’s own is not meaningful
• “seed” vs. “look” + “ed”
“deter” vs. “de” + “press” > de = pre-fix and press = root but this is not the case with deter because ter is not a verb

Question 6

Task: primed lexical decision (Lima & Pollatsek, 1983)

Answer 6

• Ppts are shown letters on a screen + they have to decide whether the presented letters are a real word or random letters > they were also primed so they could have been primed with TEA for 90 ms then show teaspoon + they say whether teaspoon is a real word OR they would be primed with TEA but shown TEGSPOON + they say whether tegspoon is a real word or not > You then measure reaction time and accuracy
• 3 conditions (if the prime overlaps w/ the target, you should recognise the target faster, more overlap = more fast)
○ Condition 1: prime was TE + target was teaspoon
○ Condition 2: prime was TEA + target was teaspoon
Condition 3: prime was TEASP + target was teaspoon

Question 7

Q2. Possible outcomes + results

Answer 7

Possible outcomes:
• If morphemes are not access units + the most important thing was how much of the prime overlaps with the target, then condition 3 would have the quickest response, then 2 then 1.
• If morphemes are access units, then condition 2 (tea) corresponds to a morpheme in the target word (teaspoon) > shared unit so response would be faster > 2 would be faster than 3 and 1
Results
• When presented with TEA (c2) before seeing teaspoon + saying whether it was a real word > results show that ppts were both faster at saying teaspoon was a real word when primed with the morpheme and they had made less errors compared to the other conditions
• This happens even though there is less overlap compared to C3 which suggests we do not see the word as one whole but access in terms of morphemes
Therefore, morphemes are also access units to the mental lexicon

Question 8

What about pseudo-affixes?

Answer 8

• Rastle et al (2004) > the broth in my brother’s brothel study > looked at pseudo-suffixes + compared processing of pseudo-affixes compared to another word which is not a pseudo-suffix but controlled for by length etc
○ Corner = pseudo-suffix because corn = stand alone word and er = usually used like farmer but a corner does not corn, it is the corner of something (corner of room)
○ Brothel = not a pseudo-suffix because broth = stand alone word but el is not usually used (don’y usually have “el” at the end of things”
• First showed corner or brothel as a prime, then they saw corn or broth then say whether it is a real word or not
Possible outcomes:
• If pseudo-affixes are not seen as separate units (corn + er) then the priming+ RT should be comparable between both conditions (corner + brothel) > don’t split corner into corn + er because then you would react faster to corn
• If pseudo-affixes are used to divide words then you would expect greater priming for corner than brothel due to the above
Result
• Results show this is the case, RT is faster for the pseudo-suffix than the non pseudo-suffix
This means that both suffixes and pseudo-suffixes are extracted as access units early in word recognition

Question 9

Q3. Are letters processed at the same time (parallel) or one at a time (serial) during word recognition?

Answer 9

• Usually the longer the word is, the longer it takes to recognise the word > this is expected if you process the letters one at a time. > long word = long RT and short word = short RT
• However, if words are processed in parallel, all at once, then the RT to recognise a long word should equal to a short word’s RT
Task: reading aloud/word-naming
• Present a word, and you pronounce it out loud as quickly as possible + measure onset RT (as soon as you begin speaking, this is the RT + accuracy is considered)
The word presented can either be a non-word (sep, snutch), low frequency word (cot, crunch) or high frequency word (box, branch)

Question 10

q3. Results

Answer 10

• Found that the more letters a non-word has, the longer it takes to pronounce that word
• High frequency words (words occurring often) showed that ppts consistently responded fast regardless of length
• Low frequency words were a bit longer than high but shorter than non-word RT, there was only a slight factor of length but this was weak
○ No word length effect for HF word
○ Weak word length effect for LF words
○ Strong word length effect for non-words
• There is a strong word length effect for non-words because these words do not exist in your mental lexicon, so you need serial grapheme-phoneme conversion (coverts letter-by-letter into sound > more letters = more time needed)
• LF words showed a weak effect but it’s not clear whether this is due to word length, could just be that long LF words have fewer similar looking words so there is less “help” from them (neighbourhood effect)
• So it is not clear whether there is a length effect for LF words
• If reading proceeds letter-by-letter, we expect to see a length effect for BOTH LF and HF words.
This means that letters must be processed in parallel > to some extent because eye-tracking research shows people use the first letters most then the last letters + middle letter in the word matter least

Question 11

Q4. Are words that share letters activated during the search of a target?

Answer 11

• When you look for a word, will you activate words which look similar to the target?
• Evidence suggests that during recognition, we do not activate just one word (target), but a set of words which are similar looking > this is called Colheart’s N: orthographic neighbourhood of a word = how many other words overlap w/ the letters in the target word
• The words which activate together are those which are coded in a similar way aka words which share similar letters or syllables
- If we use form-based priming or orthographic priming and we present a non-word like LOUP before presenting target word LOUD, will ppts be quicker to detect the word LOUD? (because the similar form + overlap is meant to pre-activate loud)

Question 12

Task: four field mask priming + results

Answer 12

• First you see a mask, then a prime very quickly for 15ms, then the target for 35ms + you have to say whether the target is an existing word or not
• The prime is usually in a different case (lowercase vs uppercase) than the target, this may be because we pay more attention to things in uppercase than lowercase? More attention-grabbing
• So prime is loup (lowercase) and target is LOUD (uppercase) > is the word in uppercase real?
Results:
• Results showed that when you presented loud as a prime and LOUD as the target, you got the maximum priming effect + max accuracy in this condition
• When presenting a similar form, loup (prime) and LOUD (target) > you still get priming but it is not as strong as the condition where prime + target were the same (Bit less strong) > better than control condition where there is no overlap
• Control condition where prime was ship and target was LOUD, there was much less priming
• So a prime that shares letters with the target leads to faster recognition > this is true for words AND non-word primes
• Words that share letters are connected in the lexicon but this is negatively > if a prime is consciously recognised + it IS a word, then you will slower to react to a similar word (e.g. proud and loud)
• But if the prime is a non-word like loup, then there is facilitation > faster to recognise loud

Question 13

Q5. How does information flow in the system? Are there feedback connections between letters + words?
Task: letter detection task

Answer 13

• Ppts were shown were shown a fixation point first then 1 of 3 conditions
○ Condition 1: word condition > shown an existing word like work
○ Condition 2: non-word condition > RWOK
○ Condition 3: single word condition > K
• Task is to say what was the last letter you saw > was it K or D?
The words shown and options given (k and d) make up a word (work and word) > was there a k or not?

Question 14

Q5. Results

Answer 14

• Results show that there was better performance for detecting the letter in a word vs a non-word > they are even better at detecting letters in a word than when a letter is by itself
This suggests that info from the word helps letter identification > there is some kind of feedback from the word level to the letter level > this is the word superiority effect

Question 15

How do models of word recognition differ?

Answer 15

Two key dimensions of difference
• How are words searched in the mental lexicon
○ Word entries are searched one at a time (in a series or serially)
○ Word entries are searched all at once (in parallel).
• How does information flow?
○ Strictly one way: letters ® words (bottom-up, left to right)
Interactive: letters « words. (can feedback)

Question 16

Model 1: Forster’s search model

Answer 16

• Steps taking you from letters to the meaning of words

1. First step is where you recognise letters > e.g. C O W
2. Need to know how to look for a word entry which corresponds to these letters > to access word entries we need access units such as the first letter or first grapheme etc… If we assume word entries are divided into different bins (words beginning w/ A in one bin, words beginning with B in another etc…)
   a. So you choose an initial unit you recognise (like first letter) then this chooses the necessary bin > several possible bins (letters, graphemes, syllables, meaning)
   b. The search for which bin is appropriate is looked for in parallel > searching for A,B,C,D bins are done in parallel till the right one is located then words are searched serially
3. Once the bin is chosen, the bins are serially searched (1 by 1) > the words within the bin are ordered by frequency (most frequent word is at the top, least frequent towards the bottom) > means you can find the most frequent ones quicker
   a. Once in the bin + are serially searching you check to see if the target matches the entries > does car match cow? No > so on until cow is matched
4. Once you access the word entry “cow”, it takes you to the master file which gives you the meaning of the word entry

Question 17

Model 2: Morton’s Logogen model

Answer 17

• This model argues that information about letters get sent to all word detectors (logogens) at once.
• Information only feeds forward from letters to words > cannot feedback/go backwards
• Each logogen has a threshold level based on e.g. frequency
• When activation passes threshold > logogen fires > word recognised

Question 18

How does the Logogen model work?

Answer 18

Essentially you can have auditory stimuli or visual stimuli which then undergoes either auditory or visual analysis depending on the stimuli type
• This analysis then feeds into the Logogen system > within the logogen system, there are many word entries (for instance; cat, art and boa), each word entry has different threshold levels > the threshold will be low for a very frequent word and higher for less frequent word
• Imagine a test tube which a line for where the threshold is sat > The more info you get about the word, the more the test tube fills up to reach the threshold, for instance: for the word cat you may receive info that it contains and A so the test tube fills up a bit, then you get info about the T and once the threshold is surpassed, the logogen fires and the word is recognised > this is the logogen system
• There is also a contextual system > mainly to account for predictability
○ If you have a predictable word in a sentence such as “I want to send a letter so in the top right hand corner, I will place a …..” (stamp) > the word stamp is predictable so you won’t get much info coming in to recognise it
○ Also seen in eye-tracking > if you see a predictable word then you will fixate for a short time or skip it
The parallel stage of the model is that when you get the visual/auditory input (e.g. cow), this will activate letters C O W which then in parallel starts activating all words which contain C O and W

Question 19

Important notes about the model

Answer 19

• Info goes up from features to letters to words in one direction
• When enough information has accumulated, a threshold is passed and the letter or word is recognised.
• If not enough information accumulates, the letter/word is not recognised. e.g. “cob” will be activated by “cow”, but not enough to be recognised.
• High-frequency words have lower thresholds for recognition – they will be recognised faster.
Newer model: logogens for different modalities: visual & auditory logogens, maybe even reading, writing, listening, speaking logogens.

Question 20

Model 3: R&M Interactive Activation & Competition Model

Answer 20

• In this model, firstly, there is an extracting of features stage where they assess whether the letter contains a upper vertical line or horizontal line etc… > These features then activate individual letters and all letters activate at once
• Once you activate these letters, there will be parallel activation of all words containing these letters
• There are within-level connections which inhibit each other > so if you start getting more and more info about the word “take”, then you will start inhibiting similar looking words like trap or trip > within-level inhibition
• There is also between-level connections which can go up and down in both directions > can have top-down + bottom-up activation > this can excite or inhibit
• This between-level connection is what is different between this and the logogen model because you can have feedback going from the word back to the letters unlike logogen model
Letter perception is stronger for letters in active words because of this feedback > due to the word superiority effect, you will recognise a word faster when it is in a word

Question 21

Model 3: R&M Interactive Activation & Competition Model

part 2

Answer 21

• Can test this model by presenting letters and seeing how long it takes the model to settle on a specific word > if you give the word work, how long does it take the model to recognize this as work + not word or worr
○ Also can see whether it activates other words and to what extent
• If you present a visual stimuli the says WORK with blotches around the letters, especially K, you will usually still be able to see it as WORK and not WORR because the visual info fits with K > so you get feedback from the word level in WORK in order to recognise the last letter K
• There is no item “WORR” in the dictionary (“R” would be a letter consistent with the features of the final position).
• “WORD” is not supported by the features that are visible in the final letter (3/4 letters supported).
• “WEAK” and “WEAR” are supported by initial and final letters, but not by the middle letters – only 2/4 letters supported.
- The graph shows that there is some activation of these words because there is some support from the target word but there is the most activation for the word WORK and NO activation for the word WORR because it doesn’t exist in the mental lexicon

Question 22

Full model of word reading

Answer 22

• Explains from reading the words on a screen to pronouncing the words
• Letters are not always pronounced in the same way > A in “sat” and “speak” are different
• We have regular + irregular words and regular words are those which are predictable based on the way the word is spelled
○ In a regular word, all letters represent the most common sound for that letter > “sat” and speak”
○ Irregular words are not predictable based on how the word is spelt such as “yacht” or “steak”
• A full model needs to be able to deal with these two words but also new kinds of words because we often encounter new words which we should be able to still pronounce
• In experiments we use “nonwords” aka nonsense words which are letter strings which aren’t real words > could be real but in reality have no meaning
• Nonwords can also be pronounceable or non-pronounceable
○ Pronounceable non word: such as chotel
○ Non-pronounceable non word: such as rpteeof

Question 23

Classic Dual Route Model

Answer 23

• Creator argues that in order to accommodate the reading of all these words mentioned, two roots are needed called the non-lexical (Mediated) root and the lexical (direct) route
• First you do early visual processing (noticing lines, colours, etc…) and then you recognise the letter or grapheme etc… > after you go down either the lexical or non-lexical route
Non-lexical (mediated) route:
○ If you follow the non-lexical route then you begin converting the letters into sounds called the Grapheme/Phoneme Conversion System + you do this on the basis of the most common pronunciation of the letter > e.g. if you see a word containing EA it will be converted into the sound EA (most common pronunciation)
○ After this you go to the Phonological Encoding system where you then pronounce the word
Lexical (direct) route:
○ Once you have recognised the letters, you go to the Orthographic Input Lexicon (a mental dictionary for written words) which activates the words corresponding to these letters
○ Next you can go to the Semantic System to access the meanings of these words
○ You can then go to the Phonological Output Lexicon which tells you how to pronounce the words
○ Then you can go to the Phonological Encoding system where you actually pronounce them
• The lexical route is the route we use for words (known) whereas the non-lexical route is what is used for non-words because novel words aren’t in our mental dictionary for us to look up > we can pronounce novel words thus using the non-lexical words
• Known words can use the direct route because we already have a representation of the words

Question 24

Evidence from neuropsychology:

Answer 24

• Neuropsychology has been important for deciding between possible full reading models.> Neuropsychology = study of patients who have suffered brain damage.
○ Question: How must the undamaged system have been arranged so that damage can produce the observed patient data?
• Neuropsychologists usually study single cases as opposed to groups because patients will rarely ever have the same damage which prevent averaging of their performance
• When studying brain damage cases it is important to look at accuracy (aka how good are they at reading words) and also errors because errors are not random
E.g. if a brain damage patient saw a cat and called it a dog then there is an underlying reason for this > this means what their performance shows is something different from another patient who says doc instead of dog when seeing dog

Question 25

Damage to the lexical system:

Answer 25

• What tasks would be difficult with this?
○ Reading of words would be difficult because you use that system to read words
○ What tasks will be normal or near normal? Reading of non-words or novel words because the non-lexical route would be used because these words use the GPC
○ What errors are produced? Impairment of irregular words because we have to through the lexical route to access irregular words > if you went through the GPC route you would pronounce it incorrectly
For regular words you can use either the lexical or non-lexical route because it is in your mental lexicon and the GPC would allow you to correctly pronounce the word anyways

Question 26

Predictions when there is damage to the lexical system:

Answer 26

Lexical route is impaired so ALL reading goes through the GPC > means words which are irregular are likely to be pronounce incorrectly
• Percent correct:
○ Would expect irregular word reading to be poor
○ (regular words are converted via grapheme-phoneme conversion and then the sound form is used to understand)
○ Non-word and regular words should be read ok
• What errors do we expect? We would expect regularisation errors which is where irregular words are being regularised
○ So instead pronouncing irregular words how they should be, you pronounce it using regular rules so instead of saying PINT you would pronounce it like the way you would with MINT (try it out loud)
○ “Flood” will sound like “mood”
If we find this, this suggests that there would be two different routes

Question 27

Evidence: Surface Dyslexia

Answer 27

• Patient KT (McCarthy & Warrington, 1986)
• When KT was asked to read irregular words (also called exception words) she performed very badly (47% on high frequency words and 26% on low frequency”
• When KT was asked to read regular words they got 100% on high frequency words and 89% on low freq words > non word reading is also
• KT also showed regularisation errors so she would say ko-lo-nel instead of colonel > this happened 85% of the time
• What is surface dyslexia? Surface dyslexia is an “Acquired” dyslexia which is a result of a stroke or brain injury
This is completely different from developmental dyslexia which does not result from brain damage

Question 28

Is there a double dissociation? Would be strong evidence for two independent routes

Answer 28

• If the non-lexical route is impaired, then reading can only occur via the lexical route
• Predictions: percent correct
○ Irregular and regular word reading would be good as they usually go through the lexical route anyways
○ Non-word reading would be bad because these words would not be available in the mental lexicon to help pronounciation
○ Low frequency word reading would also be worse than high frequency word reading because some evidence suggests that the GPC is involved in the reading of low frequency word
What errors would occur? If you try pronouncing non-words then you would expect lexicalisation errors because you cannot find the words in the mental lexicon (As it doesn’t exist) so you activate something close to it > e.g. instead of saying “plash” you may say “page”, “crash”

Question 29

Evidence: Phonological Dyslexia

Answer 29

• Patient WB (Funnell, 1983)
• WB’s word reading for regular (90%) and irregular words was good (80%) but reading for non words was 0% so he could not read non-words at all so he couldn’t read a word he hadn’t encountered before
• Non-word errors were also shown: lexicalisation errors > e.g. instead of “hean” they would say “hen” > instead of “ploon” he would say “spoon”
Phonological dyslexia: another acquired dyslexia - results from a stroke or brain injury

Question 30

Deep Dyslexia:

Answer 30

• Possibly a more severe form of phonological dyslexia
• People with deep dyslexia have trouble with reading non words, function words, abstract words (imageability effect), nouns are easier than adjectives + adjectives better than verbs + make visual errors (e.g. say think instead of thing)
• Defining effect is semantic paralexia: produces words which are related in meaning (e.g. if they saw “duel”, should produce sword or rapier which are words related to duel but not duel itself)
Why are they so many associated symptoms? Most likely because there is damage to the GPC route but also damage to the semantic system + perhaps nearby areas

Question 31

Neuropsychological data and data from intact people

Answer 31

• Data from patients strongly indicate independent capacities so two routes likely exist + are separate. Logic of double dissociation.
• If this is true, there should be evidence in people without participants > because we assume the same system produces both control + patient performance > (no guarantee that all aspects of patient performance will be seen in the undamaged brain, but we’re happier when our results converge.)
- Data from the neurotypical ppts do not seem to show the same kind of story

Question 32

Problems for the dual route model: Normal data

1. Glushko (1979)

Answer 32

• Glushko (1979) suggests that lexical effects can ALSO occur on non-words which go through the GPC route so why are there lexical effects here?
  • Tave vs Taze (non-words)
  ○ Tave has neighbours which are regular words such as save, cave, dave, nave + exception high frequency word “have”
  ○ Taze has neighbours like haze, maze, gaze, daze + no exception word as part of your logical neighbourhood
  • According to dual route model, tave should be the same performance as taze because both forms are non-words so you use the GPC. > the GPC is independent so similar words should have no influence
  • Grushkal found that non-words with an exception/inconsistent word neighbours were read slower than non-words with no exception neighbour + more errors were made (646ms + 21.7% errors for non words with an exception + 617ms and 6.2% errors for non words without an exception)
  • So you are more likely to pronounce tave as tav following pronunciation of have
  Lexical neighbours influence both pronunciation and errors for non-words so not all non-words are processed in the same way which goes against the dual route model

Question 33

Problems for the dual route model: Normal data

2. Regularity effects on reading words

Answer 33

• Finds that there are regularity effects when neurotypical people read words
• Regular consistent words where rhyming neighbours are all pronounced in the same way > e.g. “wade” rhyming neighbours jade, fade, bade
• Regular inconsistent words where rhyming neighbours are not necessarily pronounced in the same way > e.g. “wave” rhyming with “gave”, save, pave, cave but also HAVE
• Dual route model argues you should be able to say wade and wave equally as fast because both are words using the direct route + consistency of neighbours should not influence reaction time or errors
• But, regular consistent words are name faster + more accurately than regular inconsistent words and exception words > results by arrow
• So, not all words are processed in the same way