Speech Perception in Dyslexic Children With and Without Language Impairments

UCLA Working Papers in Phonetics, No. 103, 30-47



Speech Perception in Dyslexic Children With and Without Language Impairments*
Frank Manis, Dept. Psychology, USC
Patricia Keating, Dept. Linguistics, UCLA

Developmental dyslexia refers to a group of children who fail to learn to read at the normal rate
despite apparently normal vision and neurological functioning. Dyslexic children typically
manifest problems in printed word recognition and spelling, and difficulties in phonological
processing are quite common (Lyon, 1995; Rack, Snowling, & Olson, 1992; Stanovich, 1988;
Wagner & Torgesen, 1987). The phonological processing problems include, but are not limited
to difficulties in pronouncing nonsense words, poor phonemic awareness, problems in
representing phonological information in short-term memory and difficulty in rapidly retrieving
the names of familiar objects, digits and letters (Stanovich, 1988; Wagner & Torgesen, 1987;
Wolf & Bowers, 1999).

The underlying cause of phonological deficits in dyslexic children is not yet clear. One
possible source is developmentally deviant perception of speech at the phoneme level. A
number of studies have shown that dyslexics' categorizations of speech sounds are less sharp
than normal readers (Chiappe, Chiappe, & Siegel, 2001; Godfrey, Syrdal-Lasky, Millay, &
Knox, 1981; Maassen, Groenen, Crul, Assman-Hulsmans, & Gabreels, 2001; Reed, 1989;
Serniclaes, Sprenger-Charolles, Carré, & Demonet, 2001;Werker & Tees, 1987). These group
differences have appeared in tasks requiring the labeling of stimuli varying along a perceptual
continuum (such as voicing or place of articulation), as well as on speech discrimination tasks.
In two studies, there was evidence that dyslexics showed better discrimination of sounds
differing phonetically within a category boundary (Serniclaes et al, 2001; Werker & Tees, 1987),
whereas in one study, dyslexics were poorer at both within-phoneme and between phoneme
discrimination (Maassen et al, 2001). There is evidence that newborns and 6-month olds with a
familial risk for dyslexia have reduced sensitivity to speech and non-speech sounds (Molfese,
2000; Pihko, Leppanen, Eklund, Cheour, Guttorm & Lyytinen, 1999). If dyslexics are impaired
from birth in auditory processing, or more specifically in speech perception, this would affect the
development and use of phonological representations on a wide variety of tasks, most intensively
in phonological awareness and decoding.

Although differences in speech perception have been observed, it has also been noted that
the effects are often weak, small in size or shown by only some of the dyslexic subjects (Adlard
& Hazan, 1998; Brady, Shankweiler, & Mann, 1983; Elliot, Scholl, Grant, & Hammer, 1990;
Manis, McBride-Chang, Seidenberg, Keating, Doi, Munson, & Petersen (1997); Nittrouer, 1999;
Snowling, Goulandris, Bowlby, & Howell, 1986). One reason for small, or variable effects,
might be that the dyslexic population is heterogeneous, and that speech perception problems are
more common among particular subgroups of dyslexics. A specific hypothesis is that speech
perception problems are more concentrated among dyslexic children showing greater

* To appear in The Connections between Language and Reading Disabilities, ed. Hugh Catts and Alan Kamhi,
Lawrence Erlbaum Associates (2005).

---

phonological deficits. McBride-Chang (1996) reported structural equation analyses indicating
that speech perception was not directly related to word recognition among third graders.
Instead, phoneme awareness acted as a mediator for the relationship of speech perception and
word reading. She proposed that poor perception of the phoneme might impede the development
of phoneme awareness, which in turn interfered with early word decoding and word reading
development.

Evidence in support of this view was provided by Manis et al. (1997). They tested older
(age 10-14 years) dyslexic children who had serious delays in word recognition, but who varied
in the degree of deficit in phoneme awareness. About half of the sample of dyslexics fell within
the normal range for chronological age on a measure of phoneme awareness. Manis et al. (1997)
found that dyslexics with low phoneme awareness were more likely to have speech perception
deficits on a task requiring them to identify /b/ vs. /p/ on the basis of VOT. Five of the thirteen
cases with low phoneme awareness had abnormal categorical perception functions, as opposed to
only two of the twelve cases with normal phoneme awareness. Only one of 25 cases in the CA
group and three of 24 cases in the RL group showed abnormal categorical perception, and these
were minor deviations from normal compared to what was seen in the low phoneme awareness
subgroup. It is possible that past studies finding a significant group difference in speech
perception had a greater concentration of dyslexic children with problems in phonological
awareness. However, findings inconsistent with this viewpoint have been reported. Nittrouer
(1999) studied a sample of poor readers with considerable phonological difficulties, but failed to
observe deficits in auditory processing or speech perception.

Another possibility is that speech perception difficulties might be more common among
dyslexics with broader impairments in language. The selection criteria used in past studies of
dyslexia (e.g., typically scores within the normal range on a full-scale IQ test or on a short-form
of the IQ test) allow for the possibility that some dyslexics have mild to moderate language
delays. There is strong evidence that speech perception problems are implicated in children
categorized as specific language impaired (SLI) (Elliot & Hammer, 1988; Stark & Heinz, 1996;
Tallal & Stark, 1980; 1982; Thibodeau & Sussman, 1979). Many, but not all SLI children tend
to be dyslexic (Catts, Fey, Tomblin, & Zhang, 2002; Kamhi & Catts, 1986; Goulandris,
Snowling, & Walker, 2000).

The purpose of the studies described in this paper was to investigate the relationships
among reading difficulties, phonological processing, language impairments and speech
perception. We first present data from Joanisse, Manis, Keating & Seidenberg (2000), including
re-analyses of the data, as well as data from a follow-up study on the same subjects, administered
a year later.

Dyslexia and Specific Language Impairment

The specific question we address in this paper is why speech perception difficulties are
not consistently found in a majority of dyslexic children. One possibility is that they are
associated more with phonological deficits, as hypothesized by a number of investigators
(Adlard & Hazan, 1998; McBride-Chang, 1995; Manis et al., 1997). Still another is that speech
perception problems are part of broader language deficits found in some dyslexic children, as
hypothesized by investigators exploring the correlates and sequelae of specific language
impairment (Elliot & Hammer, 1988; Leonard, 1998; Tallal & Stark, 1980). A third view is that
the varying results of speech perception tasks in the dyslexic population might be due to lack of
31

---

sensitivity in the tasks. With a sufficiently sensitive task, it might be found that all or nearly all
dyslexics have a speech perception deficit (Serniclaes et al., 2001).

Phonological dyslexia is prominent in studies exploring heterogeneity within the dyslexic
population. Investigations by Castles and Coltheart (1993) and others (Boder, 1973; Stanovich,
Siegel, & Gottardo, 1997) as well in our lab (Manis, Seidenberg, Doi, McBride-Chang, &
Peterson, 1996; Manis, Seidenberg, Stallings, Joanisse, Bailey, Freedman, Curtin, & Keating,
1999) have identified children, termed phonological dyslexics, who exhibit specific phonological
impairments relative to word reading ability. This sub-sample of dyslexics, who often form the
majority of cases in a dyslexic sample, fit the profile of phonological impairments that is often
associated more generally with dyslexia (Rack, Snowling, & Olson, 1992; Stanovich, 1988;
Wagner & Torgesen, 1987). Surface or "delay" dyslexics, have phonological skills that are on a
par with their word reading skills. These children read as far below grade level as children
typically included in dyslexia samples, but their profile of reading and phoneme awareness skills
resemble those of younger normal readers.

While phonological processing problems are found in a majority of dyslexic children, it is
also the case that a number of children with dyslexia have a history of language impairments.
Research on specific language impairment has often been carried out somewhat independently of
studies of dyslexia, even though 50% or more of a sample of children manifesting language
delays in early childhood eventually meet the criteria for dyslexia in middle childhood (Catts et
al., 1994; Goulandris et al., 2000). Specific language impaired children typically exhibit normal
nonverbal intelligence, but have delayed or deficient development of inflectional morphology
and other aspects of grammar, as well as difficulties with phonological processing and aspects of
speech perception (Catts et al., 1994; Dollaghan & Campbell, 1998; Elliot & Hammer, 1988;
Leonard, 1998; Stark & Heinz, 1996; Tallal & Stark, 1980; Thibodeau & Sussman, 1979).
Evidence of deficits in phonological processing and speech perception raise the issue of the
similarity of dyslexia and SLI.

Despite the relatively independent development of the two lines of research on SLI and
dyslexia, there is evidence that dyslexia and SLI may share some characteristics, or that SLI may
be a part of one developmental pathway to dyslexia. Scarborough (1990) found that nearly 60%
of a sample of children who were deemed at risk for dyslexia because of a dyslexic family
member qualified as dyslexic at age 8. Data collected at age 2 1/2, 4 and 5 years of age indicated
that children who later became dyslexic had delays in the development of expressive
morphology, articulation, word retrieval, and phonological awareness compared to at-risk
children who did not qualify as dyslexics as well as children without a familial risk. Moreover,
the syntactic problems predicted unique variance in later word recognition scores, partialling out
the contribution of phonological awareness and other language variables. These data indicate
that language delays are a common predecessor of reading difficulties, suggesting a common
cause for both dyslexia and the language difficulties. Whether the cause could be localized in
phonological processing or more specifically in speech perception remains to be seen.

Goulandris et al (2000) followed a sample of children identified at age 4 as SLI. They
compared children with resolved SLI (n = 19), those with persistent SLI (n = 20) and a group of
dyslexic children (n = 20) at the age of 15-16 years on a battery of tasks. The dyslexics had the
same level of oral language skill (including phonological skill) as the resolved SLI children but
were lower in word and nonword reading and spelling. Dyslexics were equivalent to the
persistent SLI children in word and nonword reading, lower in spelling, and higher in reading
comprehension. Dyslexics were also higher in phonological and other language skills. The data
32

---

present a complex picture of the relationships between SLI and dyslexia. It is possible that what
are traditionally thought of as separate disorders of SLI and dyslexia are better conceptualized as
a spectrum of language and phonological processing problems that put a child at risk for reading
and language difficulties (Snowling, Gallagher, & Frith, 2003).

Identification Functions in Dyslexics and Normal Readers: Joanisse et al. (2000)

We will report the results of a study by our group (Joanisse et al., 2000) in some detail.
This was an initial study exploring the role of phonological impairments and broader language
impairments in speech perception. We divided dyslexics into three subgroups: a group with
delayed nonword reading or phoneme awareness (as measured by experimental tasks of nonword
pronunciation and phoneme deletion) relative to a reading-level comparison group (phonological
dyslexic, or PD group, n = 16), a group with delays in both phonological skill and oral language,
as measured by tests from the CELF (Semel, Wiig & Secord, 1995) and the WISC-III (Wechsler,
1992) of morphology and vocabulary, respectively (language impaired, or LI group, n = 9), and a
group whose language scores were normal for chronological age, and whose phonological skill
was within the range of the reading-level group (Delayed group, n = 23). The three dyslexic
subgroups were equally impaired in word reading (scoring on the 8th, 6th and 9th percentiles,
respectively). The PD and LI groups were quite impaired in nonword reading and phoneme
awareness, with the PD group tending to have the more severe impairment. Groups of 52
chronological-age matched normal readers (CA group) and 37 reading-level matched normal
readers (RL group) were also tested. The RL group allowed us to some extent to balance effects
of reading achievement on phonological or language variables. If dyslexics perform more poorly
than the RL group on a given measure, it can be argued that the dyslexics’ difficulties are not
simply a byproduct of low reading achievement. Subjects had to score at the 40th percentile or
higher on the Woodcock Reading Mastery Test, Word Identification subtest (Woodcock, 1989)
to qualify for the CA and RL group. In addition, the RL group was matched to the dyslexic
group as a whole for mean and range of Word Identification grade-equivalent scores. The mean
age for the dyslexic group was 8;7 (range 7;10 to 9;4), for the CA group it was 8;5 (range 7;11 to
9;3) and for the RL group it was 6;11 (range 6;1 to 8;1). Descriptive data for the groups are
shown in Table 1.

Table 1. Means and standard deviations for the identifying tasks in the Joanisse et al. (2000)
study.


GROUP


LI (n = 9) PD (n = 16) Delay (n=23)
CA (n = 52) RL (n = 37)
Woodcock Word Iden.




- Grade Equivalent
2.1 (0.3)
2.1 (0.3)
2.1 (0.2)
4.0 (0.6)
2.2 (0.4)
- Percentile
6.3 (5.9)
8.3 (6.2)
9.3 (4.4)
68.2 (16.4)
79.7 (15.5)
Nonword z-score
-0.9 (0.7)
-1.1 (0.3)
-0.1 (0.7)
2.1 (1.2)
0 (1.0)
Phon. Del. z-score
-0.9 (1.0)
-1.5 (0.6)
-0.02 (0.4)
0.7 (1.0)
0 (1.0)
WISC Vocabulary





Standard Score
5.1 (0.9)
8.1 (3.2)
9.1 (2.7)
10.2 (2.9)
11.8 (3.8)
CELF Word Structure




Standard Score
5.2 (1.0)
7.7 (1.9)
10.3 (2.9)
11.7 (2.9)
12.6 (2.3)

33

---

Joanisse et al. (2000) explored categorical perception along a VOT (/d/-/t/) continuum
(“dug”-”tug”) and a place of articulation (POA) (/p/-/k/) continuum (“spy”-“sky”). Perception of
VOT and POA contrasts has been found to be categorical in nature in past studies of speech
perception in both normal listeners and dyslexics (e.g., Godfrey et al., 1981; Liberman, 1996;
Maassen et al., 2001; Werker & Tees, 1987). For the /d/-/t/ contrast, “dug”-“tug” stimuli were
constructed by cross-splicing progressively more components of “tug” into “dug” from natural
speech. The result was a continuum of eight different VOT values ranging from 10 ms. to 80
ms. voicing lag, in roughly 10 ms. increments. The subjects heard six practice items at the
endpoints, with feedback, and simply pointed to a picture representing the correct word (a
cartoon figure digging or tugging on a rope). There were 40 experimental trials, with each point
on the continuum represented by 5 tokens, administered in random order. The /p/-/k/ contrast
was presented as a contrast between the words “spy” and “sky”. The place of articulation
contrast was created by varying the onset frequency of the second formant (F2) transition sweep
in the second consonant of the target word. This produced a continuum from the labial /p/ to the
dorsal /k/ phoneme. F2 onsets varied from 1100 to 1800 Hz in 100 ms. steps. Formant transition
duration was close to that of natural speech, 45 ms. A closure duration of 30 ms was chosen to
be long enough to produce a clear stop consonant percept, but short enough to present problems
if listeners had difficulty responding to stimuli presented at short intervals (Reed, 1989; Tallal,
1980). These stimuli were produced synthetically using the Klatt hybrid synthesizer on a PC
(Klatt, 1990) and recorded as 16-bit, 22.05 kHz digital sound files. There were six practice trials
with endpoint stimuli, and 32 experimental trials, four at each of eight F2 onset frequencies.

Stimuli were presented using a Macintosh Powerbook with 16-bit audio and an active
matrix screen. The responses were expected to conform to the S-shaped identification curves
typical of categorical perception tasks. To quantify the data, each child's categorization data was
fitted to a logistic function using the Logistic Curve Fit function in SPSS. This yielded a logistic
slope coefficient. Valid coefficients tend to be between 0 and 1.0, with higher values
representing shallower slopes. To control for positive skew, which can invalidate logistic
functions, we excluded coefficients of 1.2 or more.

We found speech perception deficits only in the LI subgroup. This group had an
identification function with a shallower slope than that of normal readers on both the VOT and
place of articulation dimension (see Figures 1 and 2 next page, which were not printed in the
original paper). The critical comparison is between each of the dyslexic subgroups and the RL
group. The only significant difference for “dug”-”tug” involved the LI and RL group, where the
LI group showed higher mean slopes, indicating a shallower slope. Likewise, the only
significant difference for “spy”-“sky” resulted from the LI group having a higher slope than the
RL group.

Inspecting the identification functions in Figures 1 and 2, the cross-over point appeared to
be similar in the LI group and the other groups, but the LI group was more likely to label clear
instances of /d/ as /t/ and vice versa, and likewise for /p/ and /k/. The findings are consistent
with broader or less distinct categories for phonemes.

However, an alternative possibility is that LI children experience generalized auditory
processing problems that affect attentiveness to subtle auditory distinctions. According to this
line of argument, the deficit is not as noticeable at intermediate values on the continuum, since
all of the children have difficulty categorizing those stimuli, but becomes apparent at the ends of
the continuum. This possibility can be addressed by administering a discrimination task using
34

---

stimuli along the same continuum. In addition, the discrimination task provides a method of
validating the subgroup distinctions in speech perception obtained for the identification task.

Figure 1. Voicing ("dug-tug") identification functions for the five groups in the study.
Voicing Identification Function
100
90
e
s
80
70
e
s
pons
60
"
R
PD
G
50
U
T
DEL
40
t
"
CA
30
RL
20
e
r
c
e
n
P
LI
10
0
10 20 30 40 50 60 70 80
Voice Onset Time (in ms)


Figure 2. Place of articulation ("spy-sky") functions for the five groups in the study.
Place of Articulation
Identification Function
100
90
s
e
80
70
pons
s
e
60
"
R
Y
50
K
40
t
"S
PD
n
e
30
DEL
r
c
CA
Pe
20
RL
10
LI
0
0
0
0
0
0
0
00
0
110 120 130 140 150 160 17
180
F2Onset Frequency (in Hz)


35

---

Speech Discrimination in Dyslexic and Normal Readers

Previous studies exploring speech discrimination in dyslexic and normal readers have
yielded an interesting mixture of results. In this task, subjects typically are given pairs of stimuli
from a VOT or place of articulation continuum, and asked to judge whether they are the same or
different. Discrimination of pairs that are different is expected to be poor for within-category
pairs (e.g., two different stimuli from the /ba/ end of the /ba/ - /da/ continuum). Discrimination
of pairs that cross a category boundary is expected to be much better. Brandt and Rosen (1980)
reported no difference between dyslexic children and CA controls for both an identification and a
discrimination task given for each of three continua, /ba/ -/da/, /da/-/ga/ or a VOT continuum.
However, as noted by Godfrey et al. (1981), the identification and discrimination functions were
slightly flatter for dyslexics. Godfrey et al. (1981) reported weaker discrimination across the
categorical boundary for /ba/ - /da/ and /da/ - /ga/ for dyslexics compared to CA controls. In
addition, dyslexics were found to discriminate better than the controls for within-category items
on the /da/ - /ga/ continuum. This finding is of particular interest, as it indicates dyslexics may
be as sensitive as normal readers to subtle differences in the phonetic values of the stimuli. An
inference can be made that dyslexics perceive the physical differences among the stimuli as well
as the control group, but their phoneme boundaries are less sharp. Godfrey et al. (1981)
classified dyslexics into dysphonetic and dyseidetic subgroups, using Boder's (1973) criteria, but
no differences in speech perception were found between these two subgroups. However, the
number of subjects in each group (11 dysphonetics, 6 dyseidetics) was fairly small.

Werker and Tees (1987) collected both identification and discrimination data. They
found that the slope of the identification function for /ba/ - /da/ was shallower in the dyslexics.
Dyslexic children performed more poorly than age-matched controls at discriminating "different"
pairs for both 1- and 2-step pairings. Group differences were larger, favoring the control group,
for cross-boundary pairs. Inspection of the figures indicates that there was a trend for dyslexics
to discriminate within-category pairs better than the controls, but only at the /ba/ end of the
continuum. The results replicate Godfrey et al.'s (1981) findings showing better within- and
poorer between-phoneme discrimination.

Maassen et al (2001) compared dyslexic children to both CA and RL control groups on a
voicing (/bak/ - /pak/) and a place of articulation (/bak/ - /dak/) continuum using both
identification and discrimination tasks. They found no differences in the mean slope for the
identification function between dyslexics and either control group on the place of articulation
continuum. Dyslexics and the RL group differed from the CA group but not each other on the
voicing continuum, with dyslexics and RLs showing shallower slopes than the CA group.
Dyslexics demonstrated a lower level of performance on the discrimination task than both
control groups for the place of articulation as well as the voicing continuum. Inspection of the
discrimination curves indicates that dyslexic-control group differences favoring the controls were
found for stimulus pairs that crossed the categorical boundary, but also for pairs that were
within-category. This study replicated Godfrey et al.'s (1981) and Werker and Tees (1987)
findings of poorer cross-phoneme boundary discrimination in dyslexics, but not their findings of
better within-category discrimination.

Serniclaes et al. (2001) utilized sine-wave analogues to speech stimuli to create a place of
articulation continuum, in order to determine whether the deficit in categorical perception was
specific to speech. The sine-wave stimuli were designed so that subjects could perceive them as
tones or as speech stimuli (/ba/ and /da/), depending on instructions. An additional set of
36

---

modulated sine-wave stimuli that sounded more like the natural speech versions of /ba/ and /da/
were utilized. Serniclaes et al. (2001) found that the sine-wave stimuli designated as “tones” to
the subjects were not perceived categorically (discrimination functions were flat for both
dyslexics and normal readers). In contrast, the identical sine-wave stimuli designated as
“speech” showed a peak for discrimination accuracy at the typical boundary for /ba/ and /da/
obtained for adult speakers of French. The third stimulus type, modulated sine-waves, were
apparently treated as even more speech-like by the children, as the peaks were steeper at the
phoneme boundary. Dyslexics showed less peaked discrimination curves, consistent with
weaker phoneme boundaries, and were better at perceiving differences within-category for the
sine-wave “speech” stimuli. A trend in this direction was found for the modulated sine-wave
speech stimuli. Serniclaes et al (2001) concluded that dyslexics' auditory discrimination is as
good as that of normal readers, but their phoneme boundaries are less sharp.

Although they did not utilize categorical perception tasks, Adlard and Hazan (1998)
contrasted dyslexics and CA and RL controls on a wide range of auditory and phoneme
discrimination tasks. They reported no overall group differences on speech and auditory
discrimination tasks. However, a subset of the dyslexics (4 out of 13) were poor at speech
discrimination, particularly when it involved pairs of words that were not only phonetically
similar (i.e., they differed by one phonetic feature), but in which the phonetic contrast was not
acoustically salient (e.g., sue/shoe, fine/vine, still/spill and smack/snack). Adlard and Hazan
(1998) found no difference between the subgroup of four dyslexics and either normal reader
control group in detecting differences among non-speech auditory stimuli. Adlard and Hazan's
(1998) findings suggest once again that only a small subgroup of dyslexics has difficulty with
speech perception.

Follow-Up Study of Speech Discrimination

In the present study, we were able to retest some of the children participating in the
Joanisse et al (2000) study 9-10 months later on speech discrimination, using the "spy" - "sky"
continuum. The children were also retested on Woodcock Word Identification (Woodcock,
1989), WISC-III Vocabulary (Wechsler, 1992), Nonword Reading and Phoneme Deletion.

The dyslexic and CA groups were all fourth graders. All dyslexic children had to score at
or below the 25th percentile on the Woodcock Word Identification Test (Woodcock, 1989) in the
retesting to qualify for the study. Criteria for classifying children as LI, PD or Delayed dyslexics
were the same as Joanisse et al. (2000). LI dyslexics scored at or below a scaled score of 6 on
both WISC-III Vocabulary and CELF Word Structure in the previous year. Their scores from
the 3rd and 4th grade on Vocabulary and for 3rd grade for CELF Word Structure are shown in
Table 2 along with the other scores from the 4th grade testing. It can be seen that the LI children
remained well below average in WISC-III Vocabulary at the second testing. The LI group
consisted of 7 of the 9 classified as LI in Joanisse et al. (2000). PD dyslexics had to score one
standard deviation or more below the original RL group (n = 37) in the previous year on either
Nonword Reading (an experimental list of 70 nonsense words) or Phoneme Deletion (an
experimental list of 24 real words and 14 nonwords). All but two of the LI dyslexics also would
have qualified as PD dyslexics. The PD group consisted of 13 of the 16 originally classified as
PD in Joanisse et al. (2000). Delayed dyslexics scored within one standard deviation of the RL
group on both Nonword Reading and Phoneme Deletion in the previous year. The delayed
subgroup consisted of 15 of the 22 classified in this group in Joanisse et al. (2000). The three
37

---

subgroups were very similar in overall word identification skill. The CA control group consisted
of 20 children selected at random from the original group of 52 children. The RL group
consisted of 10 children in second grade selected to have the same mean and range of Woodcock
Word Identification grade-equivalent scores as the dyslexics. Descriptive data for all of the
groups is shown in Table 2.

Table 2. Means and standard deviations for the identifying tasks in the discrimination study
(scores obtained in 4th grade unless otherwise indicated).


GROUP


LI (n = 7) PD (n = 13) Delay (n=15)
CA (n = 20) RL (n = 10)
(grade 4)
(grade 4)
(grade 4)
(grade 4)
(grade 2)
CELF Word Structure




Stan. Score (3rd grade) 5.2 (1.0)
7.5 (2.1)
10.3 (2.9)
12.3 (2.8)
12.6 (2.3)
WISC Vocabulary





Stan. Score (3rd grade) 5.1 (0.9)
8.2 (2.6)
9.6 (2.2)
10.0 (1.9)
11.8 (3.8)
WISC Vocabulary





Stan. Score (4th grade) 5.9 (2.3)
9.1 (2.8)
10.1 (3.2)
10.1 (2.6)
10.2 (2.1)
Woodcock Word Iden.




(4th grade)
- Grade Equivalent
2.7 (0.3)
2.6 (0.4)
2.9 (0.5)
5.2 (1.2)
3.0 (0.3)
- Percentile
8.1 (7.8)
7.2 (5.1)
12.7 (9.6)
69.8 (12.7)
82.8 (14.1)
Nonword z-score
-0.8 (1.0)
-1.1 (0.5)
-0.1 (0.7)
1.6 (0.8)
0.7 (0.7)

(4th grade)
Phon. Del. z-score
-1.1 (1.5)
-1.0 (1.0)
.1 (0.8)
1.3 (0.7)
.2 (0.7)
(4th grade)


It is apparent from the Nonword Reading and Phoneme Deletion z-scores collected at the
time of the discrimination task testing that the PD and LI groups were the only groups with a
phonological deficit (about one standard deviation below the original RL group across tasks).
The delayed group was still well within the range of the RL group and did not differ from this
group by Bonferoni-corrected t-tests on either measure. All three dyslexic groups scored
significantly below the range of the CA group on both measures (p-values all less than .001).
Other findings of note are that the PD group was intermediate in Vocabulary scores between the
LI and delayed groups. The overall group comparison on Vocabulary was significant, F (4, 61)
= 3.92, p < .01. Tukey post hoc tests revealed the only significant differences to be between the
LI group and each of the other groups (p-values all less than .025). There were no differences in
Woodcock Word Identification grade equivalent or percentile scores between the subgroups, and
none of the groups differed from the RL group on the grade-equivalent score by t-test. CAs were
higher than the other four groups on the grade-equivalent score (p-values all less than .001).

The speech discrimination task required children to judge whether stimuli along the "spy"
- "sky" (place of articulation) continuum were the same or different. The children heard two
words spaced 400 ms apart and responded "same" or "different". The words were played by a
Macintosh Powerbook computer over headphones. The word stimuli were identical to those
used in the identification task of Joanisse et al. (2000). There were six practice trials using
endpoint stimuli (2 same and 4 different). This was followed by 52 experimental trials. The
38

---

experimental trials consisted of eight "same" trials, four pairs of stimuli repeated twice each at
F2 onset frequencies of 1100, 1400, 1500 and 1800 Hz. There were 44 "different" trials.
Twenty-eight trials consisted of pairs separated by one step at each of seven points on the
continuum (1100-1200 Hz, 1200-1300 Hz, etc.). There were four repetitions of each one-step
pair, two in one order (e.g., 1100-1200) and two in the opposite order (e.g., 1200-1100). There
were sixteen trials of pairs differing by four steps on the continuum, four each at stimulus values
of 1100-1500, 1200-1600, 1300-1700 and 1400-1800 Hz. Based on the identification data, we
anticipated that the phoneme boundary would be located between 1400 and 1500 Hz. Thus,
there was one pair in the one-step set that crossed the phoneme boundary (1400-1500), and six
pairs that were within the boundary. All four pairs in the four-step set involved comparisons
across the phoneme boundary. It should be noted that there were many more actual "different"
trials than "same" trials. However, many times the children perceived stimuli differing by one
step as "same", so from the child's point of view, there was not a huge discrepancy in the number
of "same" and "different" responses.

The results are displayed separately for “same” trials (Figure 3), four-step “different”
trials (Figure 4), and one-step “different” trials (Figure 5). Performance was fairly good on the
“same" trials for all groups, except that the groups showed a dip in performance near the middle
of the continuum (i.e., on the 1400-1400 and 1500-1500 Hz items), with the LI group performing
the poorest on these items. In fact the LI group's score of 50% correct and the CA group's score
of 58% correct on the 1500-1500 item did not differ significantly from chance. F-tests
comparing the five groups at each of the four points on the continuum revealed group differences
only for the 1800-1800 Hz pairs. This appeared to be due to lower performance by the LI and to
some extent the Delayed groups relative to the other groups. However, the only pairwise
comparison to attain significance by Tukey post hoc test was the PD vs. LI comparison. The
general lack of group differences on the "same" trials indicates that the dyslexic groups
understood the task, and were able to judge pairs that were acoustically identical with roughly
the same accuracy as the control groups. The dip in performance at or near the category
boundary (1400-1500 Hz) probably reflects unstable perception of items that are intermediate on
the /p/-/k/ continuum. It makes sense that children would be more certain that pairs on the ends
of the continuum matched one another, as they should tend to encode these items most of the
time as the same word. Pairs in the middle of the continuum might sometimes be encoded as one
word and sometimes as the other, even within the same trial, resulting in more guessing or more
"different" responses.

Figure 4 shows the percentage of correct "different" responses made on four-step pairs as
a function of F2 onset frequency. These pairs should have been fairly easy to discriminate on
two grounds, the fact that they crossed the phoneme boundary, and that they were acoustically
quite distinct (i.e., F2 onset frequency differed by 4-steps on the continuum). It can be seen in
Figure 4 that all groups achieved better than 70% accuracy across all four pair types, with mean
accuracy on the 1300-1700 Hz pair exceeding 90% for all groups. There is a trend for the PD
group and the LI group to be somewhat lower in accuracy than the other groups. However, none
of the F-tests conducted for any of the four pairs revealed significant group differences. Results
for the 4-step comparisons once again illustrate that the children generally understood the task
and were able to distinguish items differing by 4-steps on the continuum. However, since all of
the items were both acoustically and phonemically distinct, it is not possible to determine
whether this performance reflected categorical perception. The one-step items made this
determination possible.
39

---