Language 7&8 - Shallow Processing

Question 1

What is shallow processing?

Answer 1

• Can a man marry his widow’s sister?
• Only 30% notice that to have a widow, the man must already be dead!
• Suggesting that people have not fully analysed the meaning of the word ‘widow’, which suggest that they have engaged in some degree of ‘shadow processing’

Question 2

Why shallow processing?

Answer 2

• The need for system tolerance
• The language that we encounter every day is full of ambiguities that we need to be able to get the gist of a message, even when it’s delivered badly
• Classic example: sports commentors have to speak so fast, they sometimes make mistakes
• Classic example: mixed metaphors – speaker mistakenly combines two metaphors without realizing
• Although the content of the speech, when fully analysed, doesn’t really make sense, we can still generally follow what the speaker is trying to communicate
• Analysed language input fully and deeply during processing, then we would struggle to interpret language input like this, and with the ambiguity and mistakes that are common in everyday speech

Question 3

Traditional models and assumptions

Answer 3

• Full lexical retrieval and integration into a fully specified syntactic structure
• E.g. Just & Carpenter (1980) on incremental interpretation: “readers interpret a word while they are fixating it, and they continue to fixate it until they have processed it as far as they can”
• We fully analyse the meaning of ever word and then fully integrate it into a fully specific grammatical structure
• MacDonald, Pearlmutter, and Seidenberg (1994): “The communicative goal of the listener can be achieved with only a partial analysis of the sentence,”
• However, they viewed “these as degenerate cases

Question 4

Shallow processing: 5 evidences

Answer 4

• Evidence to suggest that language processing isn’t always complete
  1. Incomplete semantic commitment
  2. Garden path sentences: lingering incorrect interpretations
  3. Pragmatic normalisation: misinterpretation of passive sentences
  4. Failure to detect semantic anomalies
  5. Failure to notice text changes

Question 5

Incomplete semantic commitment

Answer 5

1. Mary bought a brand-new Hitachi radio
2. It was in the Selfridge’s window
   - What is it?
   - The particular radio that’s he bought, so after she bought it, it was gone?
   - The type of radio, so after she bought it, it was still there?
3. Later, when Joan saw it, she decided too that it would be a good purchase
   - When reading the sentence, we may not be committing fully to exactly what the word refers to

Question 6

Garden path sentences

Answer 6

• While anna dressed the baby played in the crib
• Did the baby play in the crib?
• Did anna dress the baby?
• Suggest that once interpretation is ‘good enough’, people don’t bother clearing up the details
• Suggest that although they have successfully reanalysed the sentence, they have not ‘thrown away’ their initial misanalysis
• Leads to a ‘lingering misinterpretation’

Question 7

Pragmatic normalisation

Answer 7

• A breakdown of local semantic interpretation because of pragmatic override
• Who is the ‘do-er’?
• Active sentences: the dog bit the man (99% accurate), the man bit the dig (99% accurate)
• Passive sentences: the dog was bitten by the man (74% accurate), the man was bitten by the dog (88% accurate)
• Results: for active sentences, pp’s generally correctly indicate who performed the action. But in passive sentences, pp’s are less likely to get correct answer that would not fit with their world knowledge
• So, in more complex sentences, readers have a tendency to rely on their world knowledge to interpret the sentence meaning

Question 8

The survivor’s problem

Answer 8

• The survivors shouldn’t be buried anywhere, as they aren’t dead
• The core meaning of scenario-relevant words
  1. To survive: to be alive after some life-threatening event
  2. To be injured: to have some part of the body damaged as a result of some event
  3. To be wounded: to have an injury that leads to loss of blood
  4. To be maimed: to have an injury that renders a limb useless
• Results: pp’s were far better at detecting the anomaly (i.e. saying you wouldn’t bury the people, when the word ‘survivors’ were used)
• Suggest that relevant core meaning may aid detection
• Non-detector’s (answers pp’s gave): Ask the relatives of the deceased where they would prefer them to be buried, Find the country of origin of the dead, and bury them in whatever country is closest to their homeland, ship all the bodies home, Contact the next of kin of the dead, Bury the dead back in their own country, Have the relatives decide
• Words that ‘fit the context’ may be processed less deeply
• The influence of the scenario:
  a) Air crash (33% detection rate)
  b) Bicycle crash (80% detection rate)

Question 9

Failure to detect semantic anomalies

Answer 9

1. Easy-to-detect anomalies: he spreads the warm bread with socks
2. Hard-to-detect anomalies: how many animals of each kind did Moses take on the arc?
   - It was Noah that took the animals, but Moses fits well within the biblical context and so this kind of anomaly is harder to detect

Question 10

Why do people miss these anomalies?

Answer 10

1. shallow processing hypothesis
   - Anomalies are not detected because:
   a) The full meaning of the anomalous words is not retrieved
   b) Or integrated with the representation of the discourse
2. Reduced awareness hypothesis
   - The comprehension system retrieves the meaning of the anomalies and attempts to integrate the semantics of the word in question with the rest of the text
   - However, for some reason, the fact of the anomaly may not reach conscious awareness

Question 11

Eye-tracking anomaly detection

Answer 11

• Bohan & Sanford (2008) monitored people’s eye movements as thy read sentences containing hard-to-detect anomalies
• Is there system registration without conscious detection?
• Method: pp’s read scenarios such as “A North American jumbo jet was forced at gunpoint to land in Canada. The authorities negotiated with the scared and desperate hostages and calmed them down. The siege lasted for two days”
• The anomalous word would be ‘hostages’ as you don’t negotiate with hostages
• However, this is hard to spot as these words fit well within a hijacking scenario
• Compared readings times with control
• Results: no effects in first-pass readings times on hostages (pp’s didn’t immediately register the issue)
• Longer total reading times on hostages when anomaly was detected
• No difference in readings times when anomaly that went undetected so no evidence for ‘unconscious registration’
• Conclusions: detection is not immediate but slightly delayed (no early effects on HOSTAGES)
• Detection results in severe disruption
• Disruption only observed when anomalies are consciously detected

Question 12

Anomaly detection and ERPs

Answer 12

• ‘Hard-to-detect- anomaly: Child abuse cases are being reported much more frequently these days. In a recent trial, a 10-year sentence was given to the victim, but this was subsequently appealed
• Three conditions where brain electrical activity on the word, the anomaly is the word ‘victim’, but pp’s didn’t detect it
• ‘Easy-to-detect’ anomaly: Leon was the manager of a struggling record shop. Yesterday, the owner told him that he would have to think of new ways to sell more {letters/records} if he wanted to keep his job
• Letters = anomalous, records = non-anomalous control
  1. Is the way in which hard-to-detect anomalies are processed any different from how standard, easy-to-detect anomalies are processed?
  2. Is the processing of missed anomalies more consistent with the shallow processing hypothesis, or the reduced awareness hypothesis?
• Results: an N400 for easy-to-detect anomalies (e.g. sell more letters/ records resulted in larger negative spike)
• No N400 effect for hard-to-detect anomalies
• Late positive potential (LPP) effect for hard-to-detect anomalies
• But ONLY when they were actually detected

Question 13

What controls depth of processing?

Answer 13

• Are there any principled factors that control depth of processing?
• Could be random fluctuations in processing
• Could be that special devices exist in writing and speaking to maximise depth

Question 14

Focus and depth of processing

Answer 14

1. Logical subordination
2. Linguistic focus (e.g., cleft structures): It was John who was late for the party
3. Discourse focus (question set up by the text): Everyone was wondering who had arrived late
4. Attention-grabbing devices (e.g., bold, italics): John had arrived late to the party

Question 15

Logical subordination

Answer 15

• Logical subordination clearly distinguishes focal information from ‘extra’ information (TRUE OR FALSE)
• Subordinate: the liver, which is an organ found only in humans, is often damaged by heavy drinking
• Main: the liver, which is an organ often damaged by heavy drinking, is found only in humans
• Both statements contain false information that the liver is only found in humans
• Less likely to be detected when presented in subordinate than central
• This may suggest that people process information less deeply when it is part of a subordinate clause as opposed to being in the main clause

Question 16

Linguistic focus

Answer 16

1. Focus is on Moses: It was Moses who took two animals of each kind on the Ark
   - This cleft construction answers the question: WHO took the animals on the Ark?
2. Focus is on two animals: It was two animals of each kind that Moses took on the Ark
   - This cleft construction answers the question: WHAT was taken on the Ark?
   - So, people are less likely to make a mistake in the first case where the focus is on who put the animals on the Ark, rather than on how many animals had been put on
   - Text change detection: pp’s are presented with text that should be read normally. They are then shown it again, but sometimes one of words has changed
   - Pp’s should be more likely to spot a change if they have processed the text, or the target word at least, more deeply

Question 17

Discourse focus

Answer 17

• In test 1, experimenters manipulated the arrangements of words in the target sentence
• In test 2, they wanted to replicate this effect, but this time using a ‘discourse focus’
• [Focused] Everybody was wondering which man got into trouble. In fact, the man with the hat was arrested
• [Unfocused] Everybody was wondering what was going on that night. In fact, the man with the hat was arrested
• Results: they did successfully manage to replicate the results
• Pp’s noticed when the key word changes regardless of whether or not it’s in focus
  But when the change is more subtle, pp’s were less likely to spot this when the key word was in focus, suggesting that pp’s had processed it more deeply when it was in focus

Question 18

Text-change detection

Answer 18

• This method aims to discover when distinctions are not being made at some level of semantics
• Based on the ‘Granularity hypothesis’. This refers to the fineness of detail there is in a representation
• Focus increases the probability of detecting a change to a related word
• Suggests that information in focus is represented at a finer level of detail

Question 19

The use of attention-grabbing devices

Answer 19

• Depth of processing can be modulated by many attention-grabbing devices
• Attention-grabbing devices may work in a similar way to focus devices
• Anomaly detection:
  a) MOSES decided to take two animals of each kind on the Ark (86.5% detection)
  b) Moses decided to take TWO animals of each kind on the Ark (68.3% detection)
• Text change detection: Critical word presented in Italics
• Auditory change detection: Vocal stress on the critical word

Question 20

Shallow processing: evaluation

Answer 20

1. Think about the tasks used:
   - Anomaly detection
   - Text change detection (Do these encourage normal reading?)
   - ERPs or eye-tracking are sensitive when it comes to telling whether people have unconsciously detected an anomaly
2. Think about the materials used:
   - Are these common in everyday life?
   - Are they designed to trick people? (A ‘misdirection of focus’?)