Transcript Slide 1

Innovative Approaches to Displaying Words
-- The effect of segmentation on
word identification
Yu-Chi Tai,
Shun-nan Yang,
John R. Hayes, &
James Sheedy
College of Optometry Pacific University
Forest Grove, Oregon
June 3-5, 2006
How are complex words represented
in the mind?
e.g.,
delivery
government
speculation
truthfulness
quasiregular
…
The Big Debate:
How are words with complex structures processed?
• Direct encoding of the whole
word
• Parse a word into morphemes
to minimize redundancy
• Associate syllables with sounds
Syllable: Link visuals to sound
• Syllables are determined by sound.
– A single or a set of vocal sounds uttered with a single
uninterrupted articulation
– Equal or larger than a phoneme (single sound)
– knowledge of syllables is often implicit: one can
follow the rules even though one cannot state them
Morpheme: link visuals to meaning
• The smallest meaning-bearing units of a word
– Stems: core meaning units
• e.g., SUNBAKED = sun + baked
– Affixes
• e.g., farmer
faithful
enable
– Rul-ebased syntactic affixes
• e.g.,Tom’s
gives
displayed
e.g., Untruthfulness = un- + true + -th + -ful + -ness
• Stronger effect from stem
frequency than word
frequency on word
recognition (Carr and Pollatsek, 1985)
• Morphological primes affect
on lexical decision, but not
visual- or unrelated primes
(Murrell & Morten, 1974)
• cars – car
• card- - car
• book – car
• Pseudowords with
morphological stems are
harder to reject (dejuvenate )
(Taft & Forster, 1975)
Models of word recognition
The assumption
• Visually segmenting a word into units based
on these hypothesized processes would
differentially affect the accuracy and latency
of lexical access
 If one process is more critically utilized in lexical
access than the others, the corresponding
segmentation method should result in greater
benefit in lexical access.
Method
• Subjects:
54 native English speakers (age 18-40)
• Stimuli:
– Words of 7- to 13-letters with similar frequency
– All words in 12-point Consolas
– 3 segment conditions (by inserting 2 extra pixels of
spaces between segments):
• Syllable-based segmentation
• Morpheme-based segmentation
• No segmentation
• Three tasks:
– Threshold word recognition
– Within-word letter recognition
– Lexical decision
EXP 1.
Threshold word recognition
• Read aloud words presented at designated angluar sizes
decreased by increasing viewing distance
• Correct responses were transformed into logMAR to
represent the smallest visual angle for recognizing a
word
EXP 1. Result
Threshold Word Recognition is…
• best with syllable-based segmentation
• poorest with no segmentation
Segmentation effect on threshold word recognition
syllable
original
morpheme
0.000
0.020
0.040
0.060
0.080
0.100
0.120
LogMAR (logarithm of the minimum angle of resolution)
Figure 1. Effect of segmentation on threshold word recognition. Smaller logMARs indicate the word can be
resolved at smaller angular size. Outside white box = 95% CI of the estimated mean; Inner gray box = 84%
CI of the mean. Lack of overlap means statistically significant difference of α < .05 for the conditions.
EXP 2.
Within-word letter recognition
• Task: Identify which of two letters was presented
briefly on the tested location before
EXP 2. Result
• Accuracy & RT were similarly facilitated by syllablebased segmentation.
Segmentation Effect on
RT in Letter Recognition
Segmentation effect on
Accuracy of Letter Recognition
(Identify a letter embedded in word; correct responses only)
(Identify a letter embedded in word)
Syllable
Syllable
Morpheme
Morpheme
Original
Original
0.50
0.55
0.60
0.65
Accuracy
0.70
0.75
0.80
1000
1050
1100
Reaction Time (msecs)
1150
1200
EXP 3.
Lexical decision
• Task: Judge whether the presented word can be used
as a noun
EXP 3. Result
• Accuracy in lexicon decision was best for syllablebased segmentation
• No sig. difference on RT
Segmentation Effect on
RT in Letter Recognition
Accuracy in Lexicon Decision
("Is the word a noun?")
(Identify a letter embedded in word; correct responses only)
Syllable
syllable
Morpheme
morpheme
Original
original
0.60
0.65
0.70
0.75
Accuracy
0.80
0.85
1200
1300
1400
1500
1600
Reaction Time (msecs)
1700
1800
Conclusions
• For skilled native English readers, segmenting a complex
word into chunks improves threshold word recognition
(Syllable > Morpheme > Original)
• Syllable-based segmentation enhances word processing
at various levels.
 Demonstrate strong facilitation effect of syllabic
segmentation on phonological processing
• Application: An innovative approach to display words that
potentially facilitates readers’ word processing by placing
extra space between syllables.
Acknowledgement
This study was supported by the
Advance Reading Group of
Microsoft Corporation.
Future Study
Study 1: Single word recognition
• Task:
– Decide whether the masked word is a real word or non word.
• Manipulation:
– Original word
– Original word with wider spacing (less lateral interference)
– Original word with arbitry chunking (less interference &
meaningless chunking)
– Syllable-segmenetd word
– Morpheme-segmented word
– Nonword with similar shape (holistic route)
– Nonword homophones (syllable-phonological route)
– Related words with the same origins (morpheme-meaning,
semantic route)
– Unpronuncible nonword
• Measuring: RT & Accuracy
Study 2. Segmentation effect on word
identification in sentence reading
• Task: Recognize the disappearing word in the
sentence.
• Factors: syllabic-, morpheme-, no-segmented word
before disappearance
• Measurement: RT (& accuracy)
Study 3. Whole-passage reading
• 3 versions of text:
– No segmentation
– Syllable-segmented
– Morpheme-segmented
• Expected responses for easier format:
–
–
–
–
–
Shorter, fewer fixations
Longer saccades
Fewer regressions
Faster overall reading speed
Better comprehension
• Challenges (confounding factors):
– # segment (usually more for syllable-segment  more fixations?)
– Word width (longer for segmented words  more fixations?)
– Increase inter-letter spacing to maintain word width the same.
Study 4. Segmentation effect in nonword identification during reading
• Task: Identify a pseudoword in a passage
• Factor: original vs. syllable- and morpheme-segmented
target (pseudowords)
• Assumptions:
– If segmentation helps word identification, it will be slower to
identify a segmented pseudoword;
– If word identification is holistic, then there should be no
difference.
• Measurements:
– RT
– Fix#
– Fix dur