A comparison between computer and tabletop delivery of phonology therapy

Abstract

This paper reports on the development and evaluation of a software program aimed at assisting children with phonological impairment. An experimental approach was used whereby children's speech output was assessed pre- and post-therapy. Children were randomly assigned to a computer, a tabletop or a no therapy group. Those children receiving the computer therapy were exposed to an experimental software program that mirrored the tabletop activities using interactive computer games. The results showed no significant difference between any of the three groups with regard to change in speech output. These results may relate to the amount and frequency of therapy given and also to the heterogeneous nature of children included in the study. There was considerable variation in individual performance across all three groups and the data were therefore analysed to look for patterns that might predict performance. Stimulability and gender were identified as possible predictors. Female children and those who were able to produce a greater number of consonant speech sounds in isolation were more likely to make progress in their speech output. Future research might use a similar methodology to compare the therapy conditions but with a more homogenous group in terms of stimulability and using a greater intensity of intervention.

Keywords:

• Phonological disorder
• computer assisted therapy
• psycholinguistic approach
• predicting improvement
• speech impairment

Introduction

Over the last 10 years, there has been a gradual increase in the number of software titles that can be used to assist in the delivery of therapy for children with speech and language impairments. Developments in the resources available for therapists to use should be applauded but it is not necessarily safe to assume that children who receive computer-assisted therapy respond in the same way as those who receive traditional tabletop therapy. There is a need to know what, if anything, using a computer can add to the therapy process or indeed, if there are any problems created as a result of using them. As Nelson and Masterson (1999, p. 70) comment,

If SLPs [speech-language pathologists] could show that use of computers in intervention resulted in desirable changes in communicative skills over a shorter time than with traditional modes of treatment, there would be a compelling case for funding computer hardware and software purchases and providing the needed time and training for clinicians.

The right software could provide the clinician with new and stimulating tools. Use of such resources may lead to more efficient use of therapists’ time if it is found that gains in children's speech and language are made more quickly when information technology (IT) resources are used. However, it should not be assumed that economic benefits can be made or indeed that children will respond to the use of computer supported therapy in the same way as they would respond to tabletop approaches. Thus there is a need to consider in the first instance whether or not computers can be used in intervention with children who have speech and language difficulties and secondly to consider how children's response to that intervention differs from a tabletop presentation.

The focus of this study is on children with phonological impairment (PI), that is, whose speech is characterized by omissions and substitutions of phonemes in speech, in isolation from any known cause. This paper reports on a study that looked at the progress made by children with PI who received traditional tabletop therapy and compared them with others who received computer-assisted therapy and a group who received no therapy.

Using computers in speech and language therapy

There have been several investigations of use of computers in speech and language therapy/pathology (SLP) with adults with aphasia (Adrian, Gonzalez, & Buiza, 2003; Aftonomos, Steele, & Wertz, 1997; Fink, Brecher, Schwartz, & Robey, 2002; Katz & Wertz, 1992, 1997; Linebarger, Schwartz, & Kohn, 2001; Loverso, Prescott, & Selinger, 1992, Mortley, Wade, Davies, & Enderby, 2003) and with language impaired children (Merzenich et al., 1996; O'Connor & Schery, 1986; Schery & O'Connor, 1992; Tallal et al., 1996). The programs vary in how far they were intended to be used independently of an SLP and in whether or not they are effective when compared to tabletop interventions with populations of language impaired children (Cohen et al, 2005; Gillam, Crofford, Gale, & Hoffman, 2001; Hook, Macaruso & Jones, 2001; Troia & Whitney, 2003).

Despite the more recent proliferation of software, very few studies have investigated the use of software with children with PI. Lonigan et al. (2003) and Pokori, Worthington and Jamison (2004) examined the value of computers to develop phonological awareness in children who are at risk for reading problems, but we have identified only three studies which have compared the use of computers with tabletop activities with children with PI. These studies by Shriberg, Kwiatkowski and Snyder (1986, 1989, 1990) looked specifically at the responses of children with PI to a computer presentation considering different aspects of the assessment and therapy process. The first of these studies (Shriberg et al., 1986), an assessment task, found that computer or booklet presentation of the same picture naming task made no difference to children's articulation scores. However, children's ability to identify the pictures diminished in the computer format which also took longer to administer. They concluded that although the computer presentation had considerable potential, individual differences in children's cognitive abilities and affective needs would require variations in how the computer modality was used. Their second study, (Shriberg et al., 1989) compared computer and tabletop presentation of a therapy activity—production of target sounds in words and phrases. They found no significant differences in the children's level of engagement in the task, but noted the need for the child to make eye contact with the screen rather than with the clinician. Their third study (Shriberg et al., 1990) examined computer and tabletop presentation of sound elicitation activities where the provision of feedback was given either by the clinician or by the computer. Overall, the differing modalities were found to equally effective in eliciting the sound and there were no differences in their efficiency or level of engagement, although individual variation in response from the children was apparent. From these three studies Shriberg et al. concluded that computer and tabletop modes of delivering therapy are equally effective, efficient and engaging indicating that the computer can be added to the therapist's repertoire of tools to use in phonology therapy.

These three studies took place in the 1980s when the use of computers in therapy was first being considered and, though their findings were important in introducing the notion of using computers in phonological therapy, the capabilities of computers have advanced considerably since that time. Today, computers offer greater flexibility in terms of the activity carried out and the type of stimulus material used with animation, video and sound clips amongst the options available. Moreover, children's familiarity with computers and their improved accessibility means that pupils as young as three can interact directly with the computer itself. With current hardware and software performance, children can sometimes work independently with a computer, allowing the program to judge the child's performance and give feedback in an impartial manner. Given these issues, there is a need for more current research using contemporary technology to assist in the delivery of phonological therapy.

Methodological design problems

Previous investigations of phonological therapy have used a variety of methodologies and a number of different measures of change. A number of studies have used single cases where a child acts as their own control (Bryan & Howard, 1992; Dodd & Iacano, 1989; Dunn & Barron, 1982; Monahan, 1986; Vance, 1997). All of the children in the studies cited here showed improvement on at least one measure of speech output though the degree of improvement varied from minimal changes on an intelligibility scale but no observable change in percent consonants correct to a child who progressed from 0% to 100% use of the targeted consonant in spontaneous speech. However, there was considerable variation in the amount of therapy received and information on frequency and intensity of therapy was often missing. Moreover, there were differences in the measures used to monitor change. These features make it difficult to assess the relative contribution of the therapy approaches investigated in each study.

In some cases, single cases have been strengthened by the use of multiple-baselines to control for change (Eikeseth & Nesset, 2003; Harbers, Paden & Halle, 1999; Powell, 1993; Saben & Ingham, 1991; Weiner, 1981; Williams, 1991; Young, 1987). The difficulty with such designs is the degree to which intervention affects not only those sounds targeted in therapy but other non-targeted sounds as well. Dodd and Iacano (1989) and Monahan (1986) monitored change to number and/or type of processes. In both cases, change occurred to both trained and untrained words and target and non-target processes. Similarly, Powell (1993) found that, following intervention, 75% of both treated and untreated sounds were added to the child's phoneme inventory. It is difficult to be certain whether the therapy given in these studies did result in the changes or whether some other factor such as spontaneous maturation was exerting an effect.

A number of studies have used a group approach (Almost & Rosenbaum, 1998; Bowen & Cupples, 1999; Elbert, Dinnsen, Swartzlander, & Chin, 1990; Reid, Donaldson, Howell, Dean & Grieve, 1996). Results are sometimes difficult to interpret because of the use of alternating group designs, where the control group start to receive intervention halfway through the study and the original intervention group becomes the control group. This design does not allow for the possibility that the effects of therapy can continue after the time that direct therapy provision has ceased (Gillon, 2002). This situation is exacerbated where comparative group designs are used (Dodd & Bradford, 2000; Gierut, 1990, 1992; Gierut, Morrisette, Hughes, & Rowland, 1996; Gillon, 2000; Hesketh, Adams, Nightingale, & Hall, 2000; Klein, 1996; Rvachew & Nowak, 2001; Smith, Downs, & Mogford-Bevan, 1998; Wolfe, Presley, & Mesaris, 2003). In these, the effects of one type of intervention could be ongoing while a second, or in some cases third, treatment is given.

As a comparative group study, the design for the investigation reported here needed to take into account these issues and ensure that spontaneous maturation was controlled for and cross contamination of the two methods of therapy delivery did not take place.

Development of software

The overall aim of the project was to evaluate the use of computer software as a possible tool in phonology therapy. Studies of the take-up of computer software in a variety of professional contexts suggest that for software to be taken up and used, it must be consistent with the user's existing cognitive decision models (Elstein & Bordage, 1988). An investigation of therapists’ existing practices in phonology therapy was therefore undertaken (Roulstone & Wren, 2001). This revealed that a number of different models of therapy were in common use and indeed, that therapists frequently employed an eclectic approach, drawing on different models and approaches to therapy within the same child's therapy programme. A review of commercially available software (Wren, 2005) revealed that there was no one software program or indeed combination of programs available which allowed this eclectic approach that therapists used in their own practice.

As a consequence, new software was commissioned and developed in consultation with the International Centre for Digital Content at Liverpool John Moores University. The Stackhouse and Wells Psycholinguistic Framework (Stackhouse & Wells, 1997) was used as an underpinning framework to provide theoretical coherence to the choice of activities within the software. This framework considers that the breakdown in development can occur at one or more levels on the speech processing chain and that different approaches combat breakdown at each level. The aim in developing the software was not to constrain therapy to the Stackhouse and Wells approach but rather to ensure that therapists had access to a range of activities along the speech processing chain. The aim was also to provide computer activities which could be safely interpreted by non-SLP users. As a result the emphasis in the software is on phonological awareness activities rather than production or output activities, although it is possible for SLPs using the software to adapt the activities for use in production work.

An initial pilot software program with limited customisation facility was trialled with five children with PI in mainstream primary schools (Wren, 2005). Revisions to the software in terms of customisation and running were made to produce the trial program used in this study.

In the program used in this study, there were eight interactive games addressing four types of activity: phoneme detection, phoneme blending, minimal pair discrimination and rhyme awareness. The options for configuring the activities to individualize the games varied from one activity to another but were in summary:

• 19 possible phonemes for targets or contrasts;

• single sounds, non-words, real words and words in sentences;

• initial, medial and final word position;

• clusters and polysyllables;

• sound on or off/ pictures on or off;

• adjustable length of time between presentation of phonemes in phoneme blending activity; and

• number of pictures/stimulus material to select in blending activity.

Table I shows examples of how the software activities map onto the various levels on the Stackhouse and Wells framework; it provides a brief description of the related tabletop activity.

Table I.  Therapy activities.

Therapy Goal	Stackhouse & Wells (1997) Level	Computer therapy gamesThe interactive games are set in a bubble factory. An adult female voice with a UK accent provides the stimulus sounds and words on the computer	Tabletop therapyUsed a variety of games with pictures and objects
Identifying speech sounds in isolation Identifying speech sounds in non-words Sorting non-words according to initial/final sound	B. Can the child discriminate speech sounds without reference to lexical representations?	Phoneme detection: The bubble producing machine produces bubbles which make sounds in isolation e.g., /s/. Sound symbols are used to represent phonemes (e.g., /s/=snake). The child manipulates the balloon using the arrow keys on the keyboard to move the bubble into a pot containing the appropriate sound symbol. The number of pots (and therefore sound symbols) is pre-set by the clinician. For example the choices might include /s/, /d/, /t/, or perhaps only /s/ and /d/. If the child chooses the correct pot, the bubble disappears into the pot; if they choose incorrectly, the bubble pops. Each game gives the child 10 tries.	The child has pictures of the sound symbol representations on the table. The clinician makes a sound in isolation. The child places a brick on the sound symbol's representation of that phoneme. Clinician varies the number of choices available. An alternate version is where the clinician uses the target sound in initial or final position in a non-word. The child is required to respond in the same way as with the sound in isolation.
		In another variant of the game the bubbles produce non-words and the child is asked what sound is at the beginning (or at the end, depending on how the clinician sets the game). Once again the child manoeuvres the bubble into the correct sound symbol pot
Identifying speech sounds in real words	D. Can the child discriminate between real words?	Phoneme detection: In this variant of the game the bubbles produce words and the child is asked what sound is at the beginning (or at the end, depending on how the clinician sets the game). The child manoeuvres the bubble into the correct sound symbol pot	In a shopping game where a number of pictures/objects are placed on the table in front of the child, the child is asked to “buy” the items beginning or ending with a specified sound (e.g., /s/ - sock). Alternative objects chosen by clinician to illustrate alternatives to target sound
Sorting words according to initial/final sound		Rhyme awareness: The bubble machine produces two words which either rhyme or do not rhyme. The child is asked to click on a tick if the items rhyme and a cross if they don't. If they are correct the bubbles are taken into the bubble machine.	The child is asked to identify whether pairs of words rhyme or not using pictures of words.
Auditory rhyme judgement and detection
Phoneme blending with pictures Real word minimal pair discrimination	E. Are the child's phonological representations accurate?	Phoneme blending: The computer produces a word, one phoneme at a time. Clinician can vary the length of time between each phoneme. The child has to guide a space ship around a maze to find bubbles containing the picture of the word they think they have heard.	Variants on the shopping game: the child buys the objects specified by the clinician who says the word, one phoneme at a time (blending) or the clinician asks to buy one picture of a minimally paired set (minimal pair discrimination)
		Minimal pair: the child directs a bubble catcher character who must catch the bubble as instructed by the computer: the choice is from a set of minimally paired words, the specific phonemes included in the paired sets are set by the clinician
Visual rhyme judgement and detection	F. Is the child aware of the internal structure of phonological representation?	Rhyme awareness: the child is given a choice of 3 bubbles; the computer says the words as the child directs the bubble catcher over the bubbles; the child's task is to pop the bubble which does not rhyme	Child is asked to select the odd one out from a series of three pictures where two rhyme and a third does not. The child does not hear the words said aloud and is instructed not to say them aloud themselves.
Silent sorting of words according to initial/final sound		Phoneme detection: the phoneme detection game can be set so that the pictures appear without the word being presented by the computer so that the child has to use their own phonological representation of the word in order to sort the bubble pictures according to their initial/final sound	A posting game is used whereby the child is asked to post items into the box which contains the items initial sound (or final sound if sounds at the ends of words are being targeted). The child does not hear the items named and they are encouraged not to name them aloud themselves.
Phoneme blending without pictures	I. Can the child articulate real words accurately?	Phoneme blending: The phoneme blending game Is carried out as before but this time the bubbles which had previously contained pictures are empty. The child can click on each bubble to hear the word underneath it and then must remember where their target word is as they direct the space ship to the correct bubble.	The clinician uses a puppet to produce a word one phoneme at a time and the child is asked to name the word. No pictures are used.
Minimal pair production		Minimal pairs: in the bubble catcher game (see above), the child instructs the clinician as to which bubble to catch from the paired words.	Using the shopping game format described above, the child instructs the clinician which item out of a minimal pair to pick up.
Real word repetition
Non-word repetition	J. Can the child articulate speech without reference to lexical representations?	Phoneme detection: a repetition of the non-word activity at level B was carried out but this time the child wore headphones and was asked to repeat the non-word to the clinician who then operated the mouse and directed the bubble to the correct pot.	A puppet was used and the child was asked to repeat some of the puppet's own words which were non-words containing the target and contrast sounds in initial and/or final position.
Tongue placement activities	K. Does the child have adequate sound production skills?	Same strategy use for both computer and tabletop groups. Children were given verbal instructions on where to place their tongue (e.g., behind teeth for placement of /t/).

Aim of study

The aim of this study was to keep content of therapy constant and compare the two modes of delivery of therapy, tabletop and computer, in therapy for children with PI.

Specifically the questions being asked were as follows:

1. Can computer therapy be used in SLP for phonological impairment to cause change in speech output?

2. How does change to speech output following tabletop therapy compare with change made following computer therapy?

Method

Research design

A group design was used where children with PI were matched into triads and then randomly allocated to one of three conditions: computer therapy, tabletop therapy or no therapy. A “no therapy” condition was included to control for the effects of spontaneous maturation on the children's speech. The study adopted an experimental strategy in which performance post therapy was compared with that before therapy commenced. Ethics approval for the study was given on the basis that, where a child was randomly allocated to the no therapy condition, the maximum time he or she would be waiting for therapy was equivalent to the current waiting for therapy times in the areas where children were recruited. In addition, local SLPs were asked and agreed to prioritize the children assigned to the no therapy group when their involvement in the study ceased.

Participants

Local SLP departments assisted in recruiting children to the project. A total of 60 parents gave informed consent for their child to participate and were willing to withdraw their child from the local speech and language therapy service for the period of one school term. Children fulfilling the following criteria were included in the study (n = 33):

• Aged 4–8 years excluding children in the first term of their reception (first) year in school;

• English was a first language (in fact all children were monolingual);

• Normal hearing no structural or oral motor difficulties;

• Performance was equal to or above the 10^th centile on Test for Reception of Grammar (TROG, Bishop, 1983) and on the Ravens Coloured Progressive Matrices (RCPM, Raven, Raven, & Court, 1998);

• Performance on the Sounds in Words subtest of the Goldman Fristoe Test of Articulation (GFTA, Goldman & Fristoe, 2000) was at or below the 10^th centile;

• Speech sound errors suggested a PI with or without an additional phonetic disorder, i.e., demonstrating phonemic (whole sound) substitutions and omissions rather than phonetic (within sound) distortions alone;

• No direct speech and language therapy input (i.e., face-to-face with a speech and language therapist for the purpose of therapy rather than assessment) for the previous 3 months.

Assessments

Children were assessed pre-randomization (T1) and immediately after intervention (T2); the two therapy groups were also assessed 3 months post-therapy (T3). All assessments took place in the child's school. Assessment at T1 was carried out by the first author; assessments at T2 and T3 were carried out by a SLP who was blind to the child's therapy condition. Table II shows the timing of assessments.

Table II.  Timing of assessments.

Assessment/measure	Time 1	Time 2	Time 3 (for intervention groups only
GFTA “Sounds in Words” subtest	✓	✓	✓
Per cent of consonants correct (PCC)	✓	✓	✓
Speech processing tasks	✓	✓	✓
Non-word same/different task	✓
Auditory rhyme judgement task	✓
Auditory discrimination picture task	✓
Visual rhyme detection task	✓
Attention level rating scale	✓
Phoneme stimulability	✓

Notes: GFTA = Goldman-Fristoe Test of Articulation; ✓ assessment carried out at this time point.

The following measures were used:

• GFTA Sounds in Words subtest;

• Percent of consonants correct (PCC) was calculated for the GFTA Sounds in Words subtest as a whole and then separately for the category of consonants targeted in therapy (e.g., fricatives);

• Speech processing tasks (Vance, 2001; Vance, Stackhouse, & Wells, 1994);

• Attention level rating scale (ALRS adapted from DuPaul, 1990) (see Appendix A);

• Phoneme stimulability (children were asked to repeat each of the consonant phonemes of the English sound system in isolation. This gave a score showing the number of consonants imitated correctly.

Matching into triads and randomization

The prime measure used for matching children was the centile rating from the Sounds in Words subtest of the GFTA. Where more than three children had the same centile rating, the per cent consonants correct (PCC, Shriberg & Kwiatkowski, 1982) score for the GFTA sounds in words subtests was taken into account. In addition, consideration was given to the phonemes that would be potential therapy targets. Where the phonemes in a matched triad were not the same, they were developmentally equivalent, i.e., they would be acquired at or around the same time as each other in a typically developing child (Grunwell, 1985). Finally the type of errors made by each child was taken into consideration. Specifically, this was whether the errors demonstrated in their speech suggested a purely phonological impairment where whole speech sounds were substituted or omitted or whether there was an additional phonetic component.

Table III provides details on the 11 triads. In the case of triads 4 and 5, the type of speech errors demonstrated and the target phonemes for therapy were used to match the children over the GFTA centiles and PCC scores. This was because there was some disparity between the centiles and the PCC scores. Children were not matched according to age, year group or gender. However, these details are included for information. Children were recruited in two waves (indicated by their ID code as either A or B in Table II) based on the time of their initial assessment. Due to timing considerations, children could not be matched across waves. Once children had been matched into triads they were assigned at random to one of the three conditions: computer therapy, tabletop therapy or a no therapy control.

Table III.  Matching of children into triads.

Triad	Therapy condition *T,C,N	Child	Male/female	Year Group R**/1/2/3	Age (in months)	GFTA Standard Scores	GFTA centile	PCC	Type of errors: phonological (P) / phonological and phonetic (PP)	Targets for Therapy
1	N	A23	m	1	71	<40	<1	39	PP	/f, v, s, z/
1	T	A19	f	1	61	45	<1	37	PP	/d, f, s, z/
1	C	A20	m	1	69	<40	<1	25	PP	/f, v, s, z/
2	C	A15	m	2	82	42	<1	56	P	/k, , s, z/
2	T	A16	m	3	89	<40	<1	36	PP	/f, v, s, z, k, /
2	N	A11	f	1	69	55	1	55	P	/k, , s, z/
3	T	A24	m	1	61	58	2	53	PP	/f, v, s, z, k, /
3	C	A9	m	1	61	62	2	53	P	/t, d, v, s/
3	N	A18	m	1	62	60	3	51	P	/t, s, z, v/
4	T	A21	m	1	71	62	4	59	P	/f, v, s, z/
4	C	A3	m	1	61	76	9	60	P	/k, /
4	N	A7	m	1	65	77	10	65	P	/f, v, s, z/
5	C	A22	m	2	82	70	6	67	PP	/s, , t, d/
5	N	A5	m	3	94	71	6	84	PP	/s, r, t, d/
5	T	A14	m	1	62	71	7	63	PP	/s, , t, d/
6	N	B3	m	1	67	53	1	55	P	/t, d, s, z/
6	T	B26	m	2	77	44	1	61	PP	/k, , s, z/
6	C	B4	m	1	63	54	1	46	P	/f, v, s, z/
7	C	B30	m	R	54	53	1	24	PP	/f, s, t, d/
7	T	B5	m	R	52	57	1	35	PP	/f, s, t, d/
7	N	B18	m	R	50	47	1	39	PP	/t, d, f, s/
8	C	B21	f	R	52	50	1	26	P	/f, s, k, /
8	T	B16	f	R	51	52	2	41	P	/k, , s, z, v/
8	N	B1	f	R	61	61	2	55	PP	/p, t, k, s, z/
9	T	B23	m	1	67	60	3	58	PP	/s, f, k, /
9	N	B31	m	R	60	63	4	47	P	/f, s, k, /
9	C	B11	m	R	53	66	5	45	PP	/f, v, s, z, d/
10	N	B27	f	R	53	43	<1	23	PP	/f, s, k, /
10	T	B6	f	R	59	41	<1	36	P	/f, s, k, /
10	C	B14	m	1	71	41	<1	46	P	/f, s, k, /
11	C	B12	m	2	76	40	<1	45	PP	/f, s, k, /
11	T	B17	f	1	72	45	<1	58	P	/k, , z, v, d/
11	N	B22	m	1	65	<40	<1	48	PP	/p, t, k, f, s/

Notes: *T = tabletop, C = computer, N = no therapy, **R = Reception: The first year in UK schooling is referred to as “Reception”, GFTA: Goldman Fristoe Test of Articulation; PCC: percent consonants correct; TROG: Test for Reception of Grammar; RCPM: Ravens Coloured Progressive Matrices. The RCPM has standardization data for children aged from 5 years, 3 months to 11 years 8 months. The raw scores for those children who were younger than this at their eligibility assessment were compared with the standardization data for the youngest children available.

Recruitment of schools

The Special Educational Needs Co-ordinator (SENCo) at each of the schools attended by the 22 children randomized to receive therapy were approached and asked to participate in the study. They were asked to provide:

• an assistant or volunteer who could be present during the weekly therapy sessions and who could provide two follow-up sessions each week for 8 weeks (i.e., three half hour sessions per week); no training was provided in advance;

• access to an internet linked personal computer for the three sessions of therapy and follow up each week.

Interventions

Each child in the tabletop and computer therapy conditions received one 30-minute session of therapy per week for 8 weeks with the first author. The decision regarding the number and length of each session was based on local SLP provision and the desire to ensure that the therapy given in the study was consistent with that which a child would receive with standard National Health Service care. Each session was carried out jointly with the assistant or volunteer in school. This individual observed and participated in each activity to ensure that they understood the nature of the task; they were then asked to repeat the activities with the child on two other occasions each week. The following week the therapist checked with the assistant on how sessions had progressed during the preceding week. So for clarity, just one therapist provided both interventions, supported by homework-type sessions provided by assistants or volunteers in each child's school. The children in the no therapy condition received no therapy during this time. Table I provides a summary of the activities carried out in both therapy conditions.

Where a target could not be addressed in one mode of delivery, it was not targeted in the alternate mode. In the tabletop condition, therapy activities made use of printed pictures and table games such as tiddlywinks, snap, pairs and lotto. Puppets were used to support tabletop therapy in pretend activities such as a shopping game. In the computer therapy condition, all activities were delivered using the specially commissioned software Phoneme Factory (Wren & Roulstone, 2006).1 The only exception to this was where advice on tongue placement was given as a tabletop task for both modes of delivery to help elicit target sounds. This is not an option within the computer software but in both groups there were children for whom the production of particular consonants was problematic. In both therapy groups, children were encouraged to develop their metaphonological skills by reflecting on the sounds that they heard; the emphasis in therapy was on providing the child with opportunities to hear and use phonological contrasts.

For the first three therapy sessions, one or two related phonemes that had been selected as initial targets for therapy were used in each of the therapy activities. At the fourth week of therapy, either new target phonemes or a new word position for those phonemes already being targeted were introduced. In a small number of cases, where children were making good progress in therapy, a third target phoneme or word position was introduced in the sixth week of therapy. All targets were revisited in the final week of therapy. Targets were selected on the basis that they would be the next to emerge in the child's system (based on Grunwell, 1985) and that a change in that error pattern would make a contribution to the child's intelligibility.

Analysis

Before analysing the differences between each therapy condition, a series of Wilcoxon Signed Ranks tests were carried out in order to determine the degree of improvement made within each group between T1 and T2. A non-parametric test was used because of the small numbers of subjects within each therapy condition (n = 11). In order to look for differences between the therapy conditions in the speech output scores, analysis of variance (ANOVA) was carried out. This was done firstly with the GFTA standard score at T2 as the dependent variable, therapy condition as a random factor and the GFTA score at T1 as a covariate. This was repeated for the PCC measure and also then for the target category of consonants. Finally, as a post-hoc analysis, stepwise logistic regression was used to examine the relationship between baseline predictor variables as the explanatory variables and the child's improvement as the dependent variable.

Results

The children in the two intervention conditions each received between 18 and 24 out of a possible total of 24 sessions consisting of 8 with the research SLP and up to 16 with the assistant helping in the school. Reasons for sessions being missed were due to absences of the child from school or school activities such as trips and school plays.

Equivalence of triads

A one-way ANOVA with therapy condition as the factor and GFTA, PCC, age (in months), TROG centile, RCPM centile and the ALRS score as dependent variables was carried out. This showed that there were no significant differences between the groups at the point of matching into triads and randomization.

Within group changes

Table IV shows the descriptive results for the three groups at the three assessment time points on the GFTA standard scores and the PCC for the Sounds in Words subtest. A series of Wilcoxon Signed Ranks tests using PCC scores were carried out to determine the degree of improvement made within each group between each assessment time. These showed that all three groups made significant improvement at T2 when compared with T1 and that the computer group made further significant progress between T2 and T3 though the tabletop group did not.

Table IV.  Descriptive results for the three study groups on the GFTA standard score and PCC.

Group condition	Measure	T1	T2	T3
Computer n = 11	GFTA standard score mean (range)	51.2 (39–81)	61.7 (39–85)	66.8 (39–95)
	Standard deviation	13.2	14.2	19.8
	PCC mean score (range)	47.5 (22–69)	59.5 (31–80)	67.6 (35–91)
	Standard deviation	17.4	14.6	15.9
Tabletop (n = 11)	GFTA standard score mean (range)	53.2 (39–69)	62.0 (39–84)	65.6 (39–89)
	Standard deviation	9.9	17.0	18.4
	PCC mean score (range)	51.9 (34–66)	62.6 (43–84)	66.3 (41–88)
	Standard deviation	12.3	13.5	15.9
No therapy (n = 11)	GFTA standard score mean (range)	53.5 (39–74)	59.7 (39–76)
	Standard deviation	11.5	12.8
	PCC mean score (range)	51.2 (23–85)	59.9 (31–90)
	Standard deviation	15.5	16.1

Notes: GFTA = Goldman Fristoe Test of Articulation; PCC = Per cent consonants correct.

Between group changes

Comparisons between the GFTA and PCC scores at T1 and T2 showed that the computer group had made most progress (mean increase in GFTA standard score = 10.55; mean increase in PCC = 12) followed by the tabletop group (mean increases of 8.82 and 10.73 respectively) and then the no therapy group (6.28 and 8.73). So, the mean change in standard scores in the computer group is equivalent to two-thirds of a standard deviation; in the tabletop condition, the mean change was just over half a standard deviation and in the no therapy condition, the amount of change is just under half of a standard deviation. However, analysis of co-variance at T2, adjusted for baseline scores, showed that these differences were not statistically significant (GFTA: F_(2,29) = .416, p = .664; PCC: F_(2,29) = .271, p = .765) suggesting that therapy condition did not significantly affect progress in speech output at T2. Figures 1 and 2 show the variability in the difference in GFTA and PCC scores between T1 and T2. Given that eleven children were seen in each therapy condition, it would be inappropriate to ignore the outliers. If larger numbers had been seen, the effect of these outliers could arguably be less important.

Figure 1. Variability in difference scores between T1 and T2 on GFTA standard score.

Figure 2. Variability in difference scores between T1 and T2 on PCC measure.

Though similar differences in scores between the computer and tabletop groups at T3 compared with T2 were observed (mean increases for computer group = 5.09 and 8.1 for GFTA and PCC respectively; mean increases for tabletop group = 3.64 and 3.63), analysis of co-variance further showed that there was no significant difference between the scores at T3 adjusted for scores at T2 between each of the two therapy conditions (GFTA: F_(1,19) = .134, p = .718; PCC: F_(1,19) = 1.7, p = .208). The variability in the difference in speech output scores between T2 and T3 is illustrated in Figures 3 and 4. These show that between T2 and T3, there is wide variability in GFTA scores for the tabletop condition whilst the computer condition contains an outlier for both the GFTA and PCC scores.

Figure 3. Variability in difference scores between T2 and T3 on GFTA standard score.

Figure 4. Variability in difference scores between T2 and T3 on PCC measure.

This paper reports only the results from the speech output measures, that is, the GFTA standard scores and the PCC scores. Performance on the speech processing tasks is not reported here but is available from Wren (2005). In fact those results add little to the picture described in this paper in that the between group comparisons on the speech processing measures showed no significant differences between the three groups.

Analyses using PCC scores for sound category targeted in therapy

Secondary analyses were carried out to identify whether differences would be observed in only the category of sounds that had been targets for therapy. The scores for the no therapy condition were included in this as the targets identified prior to randomization were used for the analysis. PCC scores for only those sounds targeted in therapy were used to carry out these analyses, so if fricatives were targeted, the PCC score for all fricatives in the speech sample at each time was calculated.

An analysis of co-variance showed that there was no significant difference between the scores at T2 for targeted categories adjusted for scores at T1 between the three therapy conditions (target 1: F_(2,29) = .261, p = .772; target 2: F_(2,21) = 1.595; p = .227) suggesting that therapy condition did not significantly affect the extent of change to the targeted consonants. As with the overall speech output scores, there is considerable variation in the scores for targeted consonants.

Post-therapy, analysis of co-variance showed that there was no significant difference between the scores at T3 for targeted consonants adjusted for scores at T2 between the two therapy conditions (target 1: F_(1,19) = .253, p = .621; target 2: F_(1,13) = .682, p = .424) suggesting that therapy condition did not significantly affect the degree of change in the targeted consonants.

Predicting improvement in speech output at T2

The data reported so far show no significant differences between the three conditions in terms of overall speech output and targeted consonants categories. However, as noted, there was considerable individual variation within each group. Therefore a post-hoc question was added to the analysis: can factors be identified which predict improvement in speech output irrespective of therapy condition. The data were therefore analysed to see if predictions could be made as to which children would be likely to improve using measures collected at baseline.

Each child in the study was categorized to an improvers group (n = 14) or a non-improvers group (n = 19) regardless of which therapy condition they had received. An arbitrary figure of ±.5 SD of all children on the GFTA at T2 compared with T1 was used to categorize the children. Table V shows the mean changes in GFTA scores for each of these groups.

Table V.  Mean change in GFTA score between T1 and T2 for improvers and non-improvers.

	Change in GFTA between T1 and T2	Change in GFTA between T2 and T3
Improvers n=14
Mean	17.29	6.91
Range	7–40	−2–26
SD	9.98	8.22
Non-Improvers n=19
Mean	2.11	1.82
Range	−6–6	−12–21
SD	3.26	9.53

A stepwise logistic regression analysis was performed with the categorization to an improvers or non-improvers group as the dependent variable and a number of predictor variables: therapy condition, gender, age at T1, scores at T1 for each of the speech processing measures, phoneme stimulablity, type of errors, TROG and RCPM centiles at eligibility assessment and the ALRS score (Table VI). This shows that only phoneme stimulability was significant (p = .032) in predicting which children would improve, indicating that children with better stimulability made more progress, though gender was near significance with p = .063.

Table VI.  First logistic regression with all predictor variables and p values.

	B	SE	Wald	df	P	Exp(B)
Therapy condition	−.988	.728	1.841	1	.175	.372
Sex (baseline girls)	3.639	1.957	3.458	1	.063	38.056
Age (T1)	−.051	.084	.368	1	.544	.950
*NWSD (T1)	−.155	.276	.314	1	.575	.857
*ARJ (T1)	−.269	.307	.764	1	.382	.764
*ADPT (T1)	.070	.065	1.153	1	.283	1.072
*VRD (T1)	−.091	.271	.113	1	.737	.913
*Phoneme stimulability	.720	.336	4.580	1	.032	2.054
*Type of error	−.140	1.181	.014	1	.906	.869
*TROG	−.505	.766	.435	1	.510	.604
*RCPM	.667	.537	1.545	1	.214	1.948
*ALRS	.045	.167	.073	1	.788	1.046

Notes: *NWSD: Non-word same/different task; ARJ: Auditory rhyme judgement task; ADPT: Auditory discrimination picture task; VRD: Visual rhyme detection task; Phoneme stimulability: a count of the number of phonemes which could be imitated in isolation at T1 out of a total of 23 phonemes (i.e., all the phonemes of Standard English except /h/); Type of error: pure phonological or a mixed phonological and phonetic group; TROG: Test for Reception of Grammar; RCPM: Ravens Coloured Progressive Matrices; ALRS: Attention Level Rating Scale score.

The logistic regression was then repeated with just phoneme stimulability and gender as the predictor variables. This time, both gender and the phoneme stimulability measure were significant (gender: p = .045, phoneme stimulability: p = .017). Together these accounted for an increase in prediction from 57.6% to 66.7%.

Discussion

The results of this study showed that no statistically significant difference was found between the computer and tabletop group with regard to change in speech output. This in itself was unremarkable as the questions asked were with regard to whether or not computer therapy using the specially commissioned software Phoneme Factory¹ (Wren & Roulstone, 2006) could be used with children with PI and to what extent the two modes of presentation were equivalent; no assumptions were made as to whether one would be better than the other. An unexpected finding however was the lack of a significant difference between either the computer or the tabletop conditions and the no therapy group. Given that the content of the therapy was based on an approach with an evidence base in the literature, it was expected that the two groups in receipt of therapy would make significantly more progress than that of the no therapy group. The possible reasons for this finding are discussed below as questions 1 and 2 posed in the introduction are considered in turn. Given the variability in children's speech output at T2 and T3, the usefulness of the data in helping to predict change in speech output over time is then discussed. Relevant observations in the literature are also considered and the direction of future research in this field proposed.

1 Can computer therapy be used in SLP for PI to cause change in speech output?

A comparison of computer therapy with no therapy showed that though both groups made significant progress in their speech output between T1 and T2, there was no statistically significant difference between the two groups in the amount of progress made. Examination of the mean differences show that, on average, the computer group made 4 points more progress that those in the no therapy control group, that is, less than half a standard deviation. However, an analysis of the variability in the overall speech output measures though showed that some children in both groups made considerable improvement in their speech while others made little or no change.

These findings contrast with other phonological efficacy studies reported in the literature where generally a positive relationship between those receiving therapy has been found when compared to children having no therapy input. It is possible that alternative research designs used in other investigations may account in part for these findings. Many studies have used single case studies in which the therapy given has been specifically matched to an individual's presenting need (Bryan & Howard, 1992; Dodd & Iacano, 1989; Dunn & Barron, 1982; Saben & Ingham, 1991; Vance, 1997; Weiner, 1981). In contrast, in the study reported here, the intention was to keep the therapy content broadly similar for each child so as to allow for a comparison of the mode of therapy delivery.

Furthermore, where group studies have been reported in the literature, the amount and intensity of therapy has been greater than the 8 sessions of direct SLP and 16 sessions of follow-up practice offered in this study (Almost & Rosenbaum, 1998; Bowen & Cupples, 1999; Rvachew, Nowak & Cloutier, 2004). In contrast, Denne, Langdown, Pring and Roy (2005) found that a group of children receiving 12 hours of phonological awareness therapy were no different from an untreated group in terms of speech production post-therapy. Indeed, Law, Garrett and Nye (2004) identified longer duration of therapy (specified as more than eight weeks) as a potential factor in a positive outcome to therapy. The issue of timing for intervention concerns not only the amount of therapy but the overall time frame. For example one might expect older children who have had therapy previously, might continue to benefit from their previous therapy. However, those children may also be the more severe or chronic children who will require more intervention over an extended time in order to progress. None the less, in this study age was not a significant predictor of progress in the post-hoc regression analyses.

2 How does change to speech output following tabletop therapy compare with change made following computer therapy?

A comparison of the computer therapy condition with the tabletop condition showed that both groups made significant progress in their speech output scores at T2 compared with their scores at T1. At T3, only the computer therapy group showed a significant improvement in the speech output measures compared with the scores at T2. There were no statistically significant differences between the computer and tabletop therapy group however. Furthermore the differences between the average amount of change between the two groups on the GFTA standard scores was less than 2 points. Once again, results showed considerable individual variability.

Some comparative studies of phonological therapy have used larger numbers of participants in their investigations than in the study reported here (Gillon, 2000; Hesketh et al., 2000; Klein, 1996; Rvachew & Nowak, 2001; Smith et al., 1998). With nine children in each of the two therapy groups, the Smith et al. (1998) study is closest to that in this study. However, they used an alternating treatments design where children received one therapy followed by another. As Gillon (2002) has shown that gains from therapy may be observed after therapy has ceased, it is difficult to isolate the relative contributions of each therapy received in the Smith et al. study.

The largest samples were used by Gillon (2000) and Hesketh et al. (2000) with 61 children in each study. Interestingly, neither study showed a significant difference between the therapies offered to the children in each of the groups. Gillon's phonological awareness group did show a trend towards more improvement in speech output compared with a traditional therapy group and a minimal therapy group. This was following 20 hours of twice-weekly therapy over a 10-week period, i.e., more than twice as much therapy as was provided for the children in the study reported here. Hesketh et al. found their two treatment regimes, articulation therapy and metaphonological therapy, to be equivalent following 10 weekly sessions of therapy and postulate that one of the reasons may be related to the heterogeneity of the children and the lack of fit between the particularity of the child's disorder and the design of the intervention.

On the basis of this study alone, it is impossible to adequately answer questions 1 and 2. Based on the results presented here, the suggestion is that this software, Phoneme Factory¹ (Wren & Roulstone, 2006) does not produce change in speech output over and above spontaneous change but also that there is no difference in speech output when a child receives tabletop therapy compared to no therapy. However, the findings in the study reported here did suggest a trend towards greater improvement in the computer group compared with the tabletop therapy group and in turn when compared with the no therapy condition though these differences were not shown to be significant. Changes in the computer group were on average 4 points greater on the GFTA standard score than for children in the no therapy group. It is possible that with a larger sample or with a more prolonged or intense therapy regime, a significant finding may have been seen.

A more important question is whether or not the computer therapy using the software used in this study can be added to the repertoire of activities available to the SLP in phonological therapy. Given that the response of the group of children receiving computer therapy was equivalent to that of the group receiving tabletop therapy, one could argue that the software used in this study could be used as an additional resource for children with PI. In addition though, a repeat of this study with a larger sample or more intense therapy would provide further evidence on the content of the therapy given by either confirming or rejecting the trend described above as significant. In addition to this, it should be remembered that this study was investigating one software program and rather than to be used as a basis for all computer support, other software programs should also be investigated with regard to their potential contribution for children with PI. Finally it is worth noting an independent study of this software which compared the children's attention in the two modes of delivery by videotaping the second and seventh therapy sessions (Jamieson, 2004). Jamieson found that the computer modality was more engaging on two measures of attention (eye gaze and posture), but was similar to the tabletop modality on facial expression and redirections. However, Jamieson found that, in the computer modality, the therapist made more requests for the child to look at her—that is, to make eye contact or look at her face. Jamieson concluded therefore that whilst the computer approach may be useful for some children because of its high engagement value, it may be less useful if children are at a stage of therapy requiring frequent eye contact, thus confirming the findings of Shriberg et al. (1989). So, whilst the computer modality is engaging for children, clinicians will need to judge how far eye contact with the child is a necessary part of therapy for any particular child and therefore how appropriate a computer modality is at that particular point in therapy.

3 Can factors be identified which predict improvement in speech output?

In all three therapy conditions, some children made considerable progress in their speech output while others made negligible improvement. One explanation is that, in spite of attempts to recruit a homogenous group of children with phonological impairments, at least three groups of children were represented in the sample: those who made spontaneous progress without the need for therapy; those who needed therapy and then made progress, and those for whom the type of therapy offered in this study did not address their specific pattern of difficulties. Furthermore, given that the greatest amount of progress made by individual children was seen in the computer group, followed by the tabletop group and finally the no therapy group, one might argue for a fourth group—those children who would have made spontaneous progress without therapy but who made greater progress and possibly progressed more quickly when therapy was offered.

As noted above Hesketh et al. (2000) suggested that the heterogeneity of her sample was a possible reason for the lack of differentiated response to the two therapies. Similarly, Crosbie, Holm and Dodd (2005) identified two subgroups of children with PI and found that each subgroup responded more positively to different types of intervention. Therefore, if subgroups of children such as those described above can be identified, then therapy can be provided more effectively and efficiently for children with phonological impairments. Thus, there is a value in using the data to see if it can be used to predict those who will respond to the therapy offered versus those who need a different type of therapy or those who would make progress spontaneously.

The findings presented showed that two factors were significant in assisting the prediction of children who made progress. (Together they accounted for an increase in prediction from 57.6% to 66.7% suggesting that other factors are also involved.) These two factors were gender and phoneme stimulability. Female children and those who were able to produce a greater number of consonant speech sounds in isolation were more likely to make progress in their speech output between T1 and T2.

With regard to gender, the limited numbers of children, and specifically girls, make it difficult to be bold in any claims about this factor. Of the 33 children in the study, 25 were boys and just 7 were girls. A larger study would be needed to confirm this finding.

However, the predictive ability of phoneme stimulability already has a base in the literature. Powell, Elbert and Dinnsen (1991) found that non-stimulable sounds were least likely to change without direct treatment while Miccio, Elbert and Forrest (1999) observed that sounds that were stimulable underwent the most change in the absence of treatment. When stimulable sounds are exposed to phonological intervention, the rate of progress is increased (Rvachew & Nowak, 2001). Further evidence of the importance of stimulability is provided by Rvachew, Rafaat and Martin (1999) who found that the children in their study of phonological therapy did not progress with the auditory perceptual techniques that were provided until they received training in stimulability. In contrast, Gierut (1998), in her review of phonology efficacy studies, concluded that outcome was more positive where non-stimulable sounds were targeted.

The ability to imitate a sound in a phoneme stimulability task indicates that the input and output channels on the Stackhouse and Wells model (1997) are intact at least at the level of the individual sound for the particular phoneme being produced. The findings reported here suggest that if a child is unable to produce a sound in isolation, then therapy may be best directed to those sounds which he or she can produce in isolation; alternatively, that the child's stimulability for the target sound is addressed before activities which place the sound in words, either as input or output tasks, are used.

This finding could therefore be useful in the clinical setting. However, the phoneme stimulability measure as it was used in this study can only tell us which children were likely to progress, regardless of whether or not they received therapy. It does not distinguish between those who would have made spontaneous progress without the need for phonological therapy or those who need a different kind of therapy approach.

Dodd (1995, p. 55) also highlights the importance of consistency of error in her classification of subtypes of PI. Though consistency of error was not specifically measured in this study, there is some evidence from a qualitative analysis of some of the individual cases that differences in consistency could have been responsible for the variations in response to the three therapy conditions within each group (Wren, 2005).

Conclusion

This study found that, although all three groups made significant progress between T1 and T2, there were no significant differences between the groups and the difference between groups can at best be described as a trend. A comparison with other studies of phonological therapy has shown that, in most cases, a greater amount of therapy was provided than that given in the study reported here. To fit with local service provision, the quantity of therapy provided in this study was limited. The need for investigations in which children receive more therapy is recognized, as is the need to use larger samples in order to highlight differences that may not be observed in small group studies. However, the challenge is to find therapy approaches which are effective but which are also practical and relevant given current service limitations. These requirements are true for investigations of both tabletop and computer-assisted therapy.

The fact that the “no therapy” control group also made a significant improvement in speech output during the period of observation emphasizes the need for a control group in any study of phonological therapy. It is not known whether the improvement was due to spontaneous maturation or to some other factor or factors exerting an effect on the children's speech output. However, if a control group had not been included, the results may have led one to mistakenly believe that the progress seen in the other two groups was greater than would be likely to be achieved with spontaneous maturation.

Considerable variation was shown within each group raising questions about the suitability of the type of therapy offered for some of the children in the study. Group studies are useful however in helping to identify the features of a child's presentation of PI which make him or her most likely to either respond to the intervention provided or to resolve spontaneously.

Further analysis of the data found that stimulability at baseline was the best predictor of performance at T2 regardless of therapy condition. What is unknown is how stimulability interacts with therapy. In the clinical setting, should intervention be limited to those children who show poor stimulability as those who show good stimulability will make the necessary progress spontaneously? Or should those with good stimulability be prioritized for therapy as these children will make rapid progress more quickly than those with poor stimulability and the possible impact of their PI on their learning and personal interactions can be minimized? Moreover, there is a need to understand how this interacts with other factors that may be relevant in predicting improvement such as consistency of error.

To summarize then, this comparison study has raised questions regarding the targeting of the therapy given and its suitability for all children with PI. With regard to the question of delivery of therapy, a further study is needed in which the participants are more homogenous, particularly in terms of their stimulability and consistency of error. Moreover, this study would need to provide a greater amount of therapy for a larger number of children to control effectively for spontaneous maturation. Until then, there is a need to continue to develop software resources for use in therapy and a need to evaluate both these resources and the content of therapy that underpins them.

Acknowledgements

The authors would like to thank the Underwood Trust for their funding of this project. We would also like to acknowledge the help of Dr Chris Jarrold as an advisor to this project and to the members of the focus group who gave their time and expertise. Our thanks also go to Anthony Hughes and Laura Miller for their advice regarding statistical analyses. Finally, we would like to thank the children, their parents and teachers for participating in the project.

Notes

1. The study was undertaken using an experimental version of Phoneme Factory. The published version has been updated in terms of the user interface and the style of the cartoon characters.

Appendix A

Attention Level Rating Scale (adapted from DuPaul, 1990)

For each statement, circle the number in one column which best describes the child

	Not at all	Just a little	Pretty much	Very much
1. Often fidgets or squirms in seat	0	1	2	3
2. Has difficulties remaining seated	0	1	2	3
3. Is easily distracted	0	1	2	3
4. Has difficulty following instructions	0	1	2	3
5. Has difficulty sustaining attention to tasks	0	1	2	3
6. Often talks excessively	0	1	2	3
7. Often does not seem to listen	0	1	2	3