The Psychological Reality of Atomic Propositions

1
The Psychological Reality of Atomic Propositions

• Simon Dennis
• Annemarie Monck
• School of Psychology
• University of Adelaide

2
How is sentence structure represented in memory?

• list of words
• syntactic units
• propositional units
• relational structures
• Syntagmatic Paradigmatic Model (Dennis 2004,
  2005 Dennis Kintsch, in press)
• When trying to interpret a new sentence people
  retrieve similar sentences from memory and align
  these with the new sentence.
• The set of alignments is a relational
  representation of the sentence.

3
Evidence for Atomic Propositions

• Free Recall
• Cued Recall
• Probe Recognition
• Reading times and recall
• Recognition Priming
• (Kintsch 1998)

4
Recognition Priming

• Subjects study sentences
• The mausoleum that enshrined the tzar overlooked
  the square.
• At test they are given words and asked if they
  appeared in the study sentences
• Dependent measure is the recognition latency on a
  second probe word following a probe to either the
  same proposition or a different proposition
• Ratcliff McKoon (1978)

5
Results

• Probe with square then with mausoleum
• priming effect 111ms
• Probe with square then with tzar
• priming effect 91ms.
• Subjects are faster if the first probe is from
  same sentence as second
• Subjects are faster again if the first probe is
  from the same proposition as second

6
A closer look at priming data

• Priming results may not be bi-directional
• Pronominal Anaphoric Reference
• e.g. The chauffeur jammed the clutch when he
  parked the truck
• clutch -gt truck 583ms
• truck -gt clutch 564ms (equivalent to the
  within prop)
• Relative Clause
• e.g. The mausoleum that enshrined the tzar
  overlooked the square
• tzar -gt square 614ms
• square -gt tzar 553ms (equivalent to the
  within prop)
• Only coordinating conjunction was bidirectional.

7
Cloze Priming

• Subjects study sentences
• The chauffeur jammed the clutch when he parked
  the truck
• Then given Cloze items
• The ____ was jammed by the chauffeur.
• The ____ was parked by the chauffeur. (or vice
  versa)
• Dependent measures
• conditional probability of getting the second
  item correct given that the first item was
  correct
• Reaction time given that first item was correct
• Using passive items gives confidence that we are
  dealing with something other than syntactic level

8
Study One

• 2x3 Factorial Design
• First factor Order in which propositions were
  probed (within subject)
• Second factor Sentence type (within subject)
• Type 2 The caretaker poisoned the rat and a cat
  dragged the corpse (Coordinating conjunction)
• Type 4 As the temptress lowered a veil she lured
  the hero (Anaphoric reference)
• Type 5 The scout who rolled the log snapped a
  twig (Relative clause)

9
Results - Accuracy

• N 36 Order Significant F(1,35)13.177, p
  0.0009

10
Results Reaction time

• RTs were logged to reduce skew
• Order Significant F(1,35)5.802, p 0.0214

11
Discussion

• Cloze priming results in opposite order to the
  Ratcliff McKoon (1978) recognition priming
• Recognition and Cloze priming referencing
  different memory structures?
• Simple propositional model is insufficient

12
Study Two

• In study 1, proposition position was confounded
  with sentence structure.
• Relative clauses appeared in first proposition
  only.
• Referents appeared in first proposition only.
• Study two focused on just these two constructions
  and varied the structure
• Pronominal Anaphoric (Cataphoric) Reference
• As the temptress lured the hero, she lowered the
  veil.
• As she lured the hero, the temptress lowered the
  veil.
• Relative Clause
• The scout who rolled the log snapped a twig.
• The scout rolled the log that snapped a twig.

13
Relative Clause
14
Anaphoric Reference
15
Conclusions

• Anaphoric Reference differences all in referent
  with second position better
• Relative Clause differences all in matrix with
  second position better
• May be complicated by change of subject
• Current Work
• Put recognition and Cloze priming in same design
  to see if there is an interaction.