Transcript The difficulty of coercion: A response to de Almeida (Pickering et al

The difficulty of coercion:
A response to de Almeida
(Pickering et al. 2005)
Martijn van den Heuvel
The goal of the experiment
• Showing enriched composition does take place.
• Sentences that require coercion are more difficult than
ones that do not require coercion.
• Showing faults in de Almeida’s experiments and
conclusions.
Critique on de Almeida
• Mostly on Almeida’s experiment 1:
– Likely a type II error due to small itemset.
– Norming the itemset (Almeida’s coercing verbs were
longer on average).
– Easier coercion due to higher frequency of items.
– No norm for plausibility of possible words.
– Relatively high removal of observations.
Pickering’s Experiment
• Use of both de Almeida’s stimuli and a new set of
normed stimuli.
– If the coerced condition is no harder than the rest in both sets,
processing difficulty is weakened
– If the coerced condition is harder than the rest, de Almeida
produced a type II error.
• Use an additional control condition:
– Full verb phrase: The secretary began to type the memo
Pickering’s Experiment: Method
•
•
•
•
•
20 participants
21 sets of stimuli (de Almeida)
33 sets of normed stimuli (Davis)
Eye-tracking
ANOVA for statistical analysis
Example set:
-
The carpenter began the table during the morning break (coerced)
The carpenter built the table ….. (preferred)
The carpenter sanded the table …..(non-preffered)
The carpenter began building the table ….. (full-VP/prefferred)
Pickering’s Experiment: Results
• Scoring regions:
– Main verb region
– Noun region
– Post-noun region
• The carpenter began the table during the morning break.
• First pass time and total time.
Pickering’s Experiment: Results
Pickering’s Experiment: Discussion
• Concludes coercion produces processing
difficulty.
– Emerging early in processing (clearly in first pass in noun region
on Davis set (?))
• Results from de Almeida set are less stable than
the Davis results  type II error
• Preferred and non-preferred conditions do not
differ if plausibility is matched (error in McElree?)
• Coercion costs emerge very early in processing
– De Almeida:“[emerge in] later interpretative
processes”
Critical evaluation
• Positive points:
– Compatible with both decompositional and
atomistic view of lexical representation.
– Use of materials and stimuli
• Negative points:
– Not all results significant
• Enlarge dataset/number of participants
– Contradictory conclusions?
Additional research
• Expectation instead of plausibility
– Specific fields render different plausibility
• Needs specific participants/stimuli
• Try to locate the involved brain parts which
are responsible for the ‘cost’
– Use fMRI if possible
• Results could reinforce the current paper’s
hypotheses and conclusions.
Questions?