Assessing Childrens Inference Generation:

1
Assessing Childrens Inference Generation What
do tests of reading comprehension
measure? Claudine Crane, Dept of Psychology,
University of York

• Introduction
• Poor comprehenders typically exhibit poor
  inferential skills and show difficulty in the
  following areas
• integrating information across sentences
  (Oakhill, 1982)
• instantiating specific nouns (Oakhill, 1983)
• using lexical cues (Oakhill Yuill, 1986 Yuill
  Oakhill, 1988)
• answering questions based on inferred information
  (Oakhill, 1984 Cain Oakhill, 1999)
• An important research question is whether
  children with comprehension difficulties show
  general difficulties with inference generation or
  whether their difficulties are more specific.

Results
Comprehension
Results
plt.05
Plt.01
plt.05
Plt.05

• Study 1 Investigating inference generation in
  different contexts
• The aim of the study was to examine the amount of
  inference generation required to answer
  comprehension questions in a narrative versus an
  expository comprehension task.
• Two comprehension measures Narrative Task and
  Expository Task
• Adult raters asked to rate each question
  according to the type of information needed to
  generate an answer
• Inference types based on the constructionist
  theory of inference generation
• Certain inferences are necessary for textual
  coherence while others serve to enrich the text
  representation
• Classify each question as
• Coherence necessary, knowledge-based inference
• Cohesive necessary, text-based inference
• Elaborativeenriching, knowledge-based inference
• Literal information explicit in the text
• Vocabulary Dependent information requiring
  knowledge of specific vocabulary

• Study 2 On-line inference generation in Poor
  Comprehenders
• Investigate inference processing skills of poor
  comprehenders.
• Use on-line measure of inference generation
• Concentrate on two specific inference types
  causal and static

Response Time
Accuracy

• RA Controls
• Less-Skilled

Main Effect Inference Type plt.05
Main Effects Inference Type plt.01 Group
plt.05 Interaction plt.05

• Method
• John and Francis watched as the bear approached
  the girls. The bear was growling and showing its
  teeth. The girls were terrified but were frozen
  with fear. A bullet struck the bear's ear and it
  reared up on its hind legs in pain. It let out a
  huge howl.
• (500 msec)
• Target sentence 1 The bear had huge paws.
  (static)
• Cathie began to scream. The bear was moving
  slowly, getting closer and closer. Now that it
  was in pain it was even more dangerous. John
  know he had to act fast. He quickly raised his
  rifle to his shoulder and fired once. The bear
  fell to the ground and did not move.
• (500 msec)
• Target sentence 2 John shot the bear dead.
  (causal)
• Comprehension question 1 Where did the bullet
  hit the bear?
• Comprehension question 2 What did John do?

Results
Participants Significant gp difference plt.05

• Conclusions
• Less-skilled comprehenders
• have difficulty with the application of
  real-world knowledge to texts
• Impairs inference generation
• Poor comprehenders showed typical pattern of
  inference generation in on-line task
• Difficulties not with process of inference
  generation per se
• Measures of reading comprehension do not always
  tap same skills
• Selection of poor comprehenders commonly done
  using Neale Analysis of Reading Ability (NARA
  II).
• Large proportion of questions require inference
  generation
• Wider Issues
• Poor Comprehenders form a heterogenous group
  (i.e Cain et al, 2004 Nation et al, 2002 Nation
  et al, 2004)
• Cannot be reliably selected using one measure of
  comprehension
• Comprehension difficulties should be viewed
  within the wider context of individual childrens
  cognitive and linguistic skills

• Both tests of reading comprehension given to a
  large sample of children.
• Calculated predicted Comprehension Scores on
  basis of Accuracy Scores on the Narrative Task
• Selected 20 children
• 10 with highest positive discrepancy between
  actual and predicted comprehension scores
• 10 with highest negative discrepancy
• Compared overall performance on both tests and
  all question types

Participants ( significant gp difference plt.05)
a standard score b scaled score