home index by author index by subject

Talk; Good Practice - Secondary

Friday stream 3 Session 11.35 - 12.50 Length 25 minutes

Improvements in Literacy: The Effect of Voice Recognition Software and Text-to-Speech on English and Dutch Dyslexic Students

Andi Sanderson and Anneke Smits

(1) De Montfort University, Lincoln, UK (2) Windesheim School of Higher Education, Zwolle, The Netherlands andi_sanderson@yahoo.com

Abstract

This paper will outline a two phase research project undertaken in the UK and in the Netherlands using Voice Recognition together with text-to-speech (TextHelp) software and measuring improvements in literacy. Recent literature reports positive outcomes with regard the use of voice recognition in both the UK and America. (Miles, Martin & Owen, 1998, Miles 2000 and Raskind & Higgins 1999). The research sought to replicate these positive improvements in literacy via the use of voice recognition software (VRS) as reported in these studies. However, in addition to using VRS, the research project also sought to isolate the precise affects of voice recognition on literacy, by including a text to speech programme (TextHelp) the outcomes of which were directly compared. Second language learning for dyslexic students is an important issue in the Netherlands. Thus the research also investigated the effects of both voice recognition software and TextHelp on the acquisition of English as a second language by Dutch dyslexic students. The paper will lead those in attendance through the research findings, with an emphasis on the following three comparisons: Comparisons of the difficulties and benefits experienced by English dyslexics using VRS to produce English text, with Dutch dyslexics using VRS to produce Dutch text. In exploring difficulties experienced emphasis will be placed on the nature of language acquisition noting the phonic regularity of the Dutch language, and how this affects VRS or indeed otherwise. Comparisons of the difficulties and benefits experienced by Dutch dyslexics using VRS to produce English text, with the difficulties and benefits experienced by English dyslexics. Comparisons of the difficulties and benefits experienced by Dutch and English dyslexics using VRS, with the difficulties and benefits using a text-to-speech (TextHelp) program.

Background

This paper reports brief findings from the first of a three phase research project exploring voice recognition software and associated improvements on literacy. A full account of all three phases of this research project will be submitted for publication in the near future. This paper therefore represents a fraction of the findings and should be read in that context.

Voice Recognition Software (VRS) has been available in the United Kingdom for some ten years. It has assisted dyslexics in secondary, Further and Higher Education to overcome some of their specific dyslexic struggles, especially those associated with spelling and handwriting (Sanderson, 1999). Moreover, recent research by Miles et al (1998, 2000) and Raskind & Higgins (1999) has shown that the use of voice recognition can have beneficial effects in improving reading and spelling ages of school aged dyslexics, and in some cases phonological processing. This research explored two aspects:

• firstly to establish and quantify the type of improvements dyslexics experience in using VRS
• and secondly, to draw conclusions about the nature of the improvements by reflecting on the two language sets, in this instance Dutch and English.

These languages were selected on the basis of their innate differences i.e. Dutch can be considered a transparent language and English as non transparent.

Voice Recognition Software

The software allows the user to input data by merely speaking into a microphone, dispensing with the need for the key board almost altogether. The software facilitates the input of data by transcribing the 'voice' into 'words' which appear on the screen in almost the same instance as they are spoken. VRS also allows the user to open and close programmes, create, edit, save and retrieve files, by using just the spoken voice in conjunction with a microphone connected to the PC. However, despite the many advantages the software offers dyslexics there are disadvantages. For example, the enrolment, training and usage of voice recognition can be problematical.

The Research

The research took place in two secondary schools, one a comprehensive in Eastwood, Nottinghamshire and the other a school in Zwolle, in the Netherlands. For the purposes of the research the English group used an English version of Dragon Naturally Speaking Preferred (Version 5) whilst the Dutch group used a Dutch version of Dragon Naturally Speaking Preferred (Version 5).

Students were first selected, then matched to a control partner before establishing the base line through a series of tests, prior to using VRS. The process involved the identification of English and Dutch students based on three criteria, age, (between 12 and 15), dyslexia (formally assessed) and native speakers (of English and Dutch respectively). From these two groups students were matched to form participant groups and control groups. Individuals were matched as closely as possible using the following criteria:-

• Reading age (word recognition test)
• Prognosis
• Silent reading rate (in words per minute)/comprehension
• Amount and type of dyslexia support
• Gender
• Non verbal reasoning (Ravens)

A control group was regarded as essential in this research as it had been noted that the Miles studies (1998 and 2000) had not included a control group and hence any improvements in the participant's literacy could have been due to other factors, such as test learning effects. By including a control group (matched to the participant group) it was envisaged that it would be possible to more effectively isolate the impact of voice recognition software this being an apparent variable between the two groups.

Once the control and participant groups were formed, base line tests to establish the level of their abilities against which any literacy progress could be measured were completed. Tests undertaken were as follows:-

• English participants and the control group completed spelling, reading and free writing tests in English.
• Dutch participants and the control group completed spelling, reading and free writing tests in Dutch.

For a breakdown of the actual tests used see Appendix 1.

Once the starting point for all students involved in the research had been established, the ten participants were then trained in the use of the software. This involved receiving approximately 30 minute training every day for five consecutive days, and was limited to basic voice input and correction by mouse. It was decided to limit training in this way to ensure those who were not so computer literate would nonetheless still command a good understanding of the software. Before starting to use VRS for their own work participants were asked to dictate a set piece of 100 words to establish the number of voice recognition errors (and type) made. Although some guidance was given to students for the first few independent sessions, they were then encouraged to use VRS for their own purposes, usually homework. Participants were asked to use VRS at least twice a week and for sessions of a minimum of 25 minutes and a maximum of 45 minutes, so by the end of the 7 week period they had all completed 9 hours of voice recognition. During this period students were interviewed fortnightly to ascertain how they were progressing. These Interviews were approximately 20 minutes and also included a brief observation of students using the software. It was during these periods that additional support was given if students were experiencing any particular difficulties. On completion of this stage all students, control and participants were asked to complete the same set of tests as at the start (see Appendix 1 for base line tests) to establish if there had been any improvement which could be attributed to the voice recognition software.

Findings

Overall it is clear from the findings that the use of VRS has not only helped students overcome many dyslexic difficulties but it has also improved certain aspects of literacy. In broad terms, results have fallen along the language divide with Dutch students making more progress with spelling and their English counterparts able to produce more in terms of free writing. For ease of understanding findings are presented here as comparative tables between the two language groups and by test.

· Reading

A number of tests were used to assess improvements in reading, a single word reading test, a single nonsense word reading test, a reading miscue and a silent reading test. The data from these four tests is presented in turn below.

Single word reading: reading age.

The table below draws from data recorded at the start of the research (base line) and at the end of the period for both the Dutch and English students, control and participant groups alike. Data is recorded in months and improvements or otherwise are indicated by a plus or minus sign. Figures vary between the two language groups as different tests were used to measure reading age. Despite these differences it is of note that overall both the Dutch and the English participant groups have made more progress than their control partners. On average the Dutch participant group improved by 0.2 months and the English group by 8 months.

Single word reading: reading age.

Single word reading: nonsense words

In a second test of single word reading, nonsense words were used, and the data is presented in the table below. In a similar fashion data here draws on both base line and post research figures. Once more improvements in both the participant groups is noted with the Dutch participant group in particular showing a marked improvement by some 10.6 words.

Single word reading: nonsense words

Reading Miscue

Data from the reading miscue is given in the table below and is once more divided into language groups. The actual data differs between the two language groups with the Dutch data giving the change in the number of words (i.e. words per minute - wpm) and increase or otherwise in the miscue, whilst the English data is a composite figure of these two variables. It should be noted with regard to the Dutch data that a positive result would be reflected in an increase in words and a decrease in miscues. As can be seen from the average figures given at the bottom of the table both the English and Dutch participant groups improved on reading miscue by a significant margin in comparison with their controls. The English participant group reduced their miscues by 8.1% compared with the control's 2.1% and almost 400% improvement over the control group. The Dutch participant group also evidenced improvements both in terms of increasing reading speed i.e. increasing wpm, and also reducing miscues. In both instances the Dutch participant group improved both in terms of wpm (by 235%) and reduction of miscues by approximately 175% in comparison with their control.

Reading Miscue

Silent reading

Data shows all four groups together with base line and post research data. It is evident from these results that both the Dutch and English participants have improved, but the Dutch have made more progress (the variance being the larger). However, the English participant group has also made significant improvements able to read on average 32.2 words more per minute, compared with the Dutch improvement of 12.8 words.

Silent reading

· Spelling Age

Despite the improvements across the board with reading for both Dutch and English participant groups, the same improvement was not reflected in spelling. The Dutch participant group did make improvements in their spelling and on average improved by 3.4 months compared with their control who improved 1.2 months. However, the English results did not mirror the Dutch improvements, and as can be seen the control group actually made more improvements than the participant group in this instance.

Spelling age (words)

· Free Writing

The results for free writing, which are reproduced in the table below, are less clear cut. In addition to the data for all four groups, figures are also given showing the number of words written and the percentage miscue within those words. Hence an all round improvement would be reflected in an increase in words and a decrease in miscues. The average data, found at the bottom of the table, shows the Dutch participant group reduced the number of miscues made but decreased in the number of words produced; whilst the Dutch control group increased the number of words written and also reduced their miscues by a very similar percentage. On these figures it could be argued that the Dutch control group actually made more improvement, than the participant group. However, of interest is the English groups, where the participant group produced on average 15.4 words with 2.7% less miscues, compared with their control who produced 4.2 additional words on average with a reduction of 0.7% in miscues. These figures indicate that the English participant group improved during the period of the research. From these findings a clear pattern of improvement is evidenced, in terms of reading for both the Dutch and English participant groups, of spelling for the Dutch participant group and in free writing for the English participant group. However, it should be noted that the English control group produced better figures for spelling as did the Dutch control group for free writing.

Free writing

Discussion

These findings are broadly in line with the results that Miles reported in both his studies (1998, 2000) as presented in the table below:-

Likewise Raskind and Higgins in their study also found improvements in participant group in comparison with their control group. Their findings are presented here showing percentage increases based on absolute scores:-

From these findings it is evident that the use of voice recognition software in addition to overcoming dyslexic difficulties for example, handwriting and sequencing of the keyboard, does also have a positive effect on literacy. This positive effect may be in consequence of either one or two enforced processes that the use of the software requires. These are briefly discussed below:-

1. Enforced Reading

In order to use voice recognition it is necessary that each word is checked for its accuracy as it appears on the screen, i.e. that the word on the screen is the intended word. If the word is incorrect then the user has to select the intended word from a menu box which displays similar sounding (and in some cases similar looking words). In undertaking this process the user is required not only to read accurately, but to also pay particular attention to discriminating between words i.e. making the phoneme/grapheme relationship specific. Monitoring and correcting words in this way forces the user to focus entirely on decoding in a manner that makes composition of words explicit. Thus literacy and reading in particular may be improved as a consequence of the software requiring the user to practice reading/decoding in such an explicit and purposeful manner.

2. Simultaneous reinforcement

A second explanation of why voice recognition improves literacy may be in consequence of the simultaneous appearance of words on the screen almost at the same instance as they are spoken. In addition the words that appear are also correctly spelt. Thus the user experiences an auditory and visual representation of each word simultaneously i.e. the phoneme/grapheme relationship. In addition it has been argued by Raskind & Higgins that the process of using voice recognition is also kinaesthetic as the user has to physically speak the words. Thus the process can be seen as multi-sensory and as such serves to reinforce learning of grapheme and phoneme relationships. This process would help dyslexics in particular recognise word outlines and word spellings, i.e. improve decoding and encoding.

From the qualitative (observed) data collected during this research it is evident that the Dutch participants used the correction box following a voice recognition miscue much more intently than their English counterparts. Thus their focus was directed to the identification of errors and the correct encoding of the word required. In consequence the Dutch participants were able to better encode single words and this may account for their better improvement with regard to spelling. This explanation concurs with the enforced reading argument outlined above. Indeed one comment from a Dutch participant also supports this notion, the participant stated:-

'I have to look more closely at my work as I have to correct my own mistakes'

However, the multi-sensory effects of using VRS cannot be discounted as their effects have been extensively documented, for example Guyer & Sabatino (1989), Foorman et al (1991) and Oakland el al (1998). In basic terms a multi-sensory approach,as Gillingham & Stillman stated as early as 1956, is:-

'based upon the constant use of all of the following: how a letter or word looks, how it sounds, how the speech organs or the hand in writing feels when producing it'

Thus based on this explanation, multi-sensory effects must be impacting on the dyslexic user and in consequence improving encoding and decoding skills.

In this regard both 'enforced reading' and 'simultaneous reinforcement' will in varying degrees positively effect and develop dyslexics ability to encode and decode more effectively.

However, what is not explained using this rationale are the differences between the two language sets and to what extent multi sensory aspects have an effect. It is clear that something more is happening here and identifying these variables could provide some explanation. Phase 2 and phase 3 of this research project address these issues.

The second phase will explore the use of text to speech software in English (Texthelp). Using Texthelp will help to unravel possible effects of text to speech (indicating spelling errors) versus speech to text (always showing perfect spelling). In some way this will indicate whether the process of actually articulating speech, hearing it and seeing it has more impact than seeing and hearing text read.

The third phase will seek to find clues as to the impact of orthography on literacy improvements affected by the use of VRS. To do this Dutch dyslexic students are provided with an English version of the VRS software and matched as before to a control group on an individual basis. If progress is made to a greater extent in free writing than spelling by this group it may indicate that the actual orthography of the English language, if not also Dutch by inference, also has an interactive effect, and may explain in some part the results evidenced in the first phase.

The results of the second and third phases of this research project will be submitted for publication in the near future to the International Journal of Dyslexia.

Bibliography

Foorman, B. R., Francis, D. J., Novy, D. M. & Liberman, D. (1991) How letter-sound instruction mediates progress in first grade reading and spelling, Journal of Educational Psychology 83: 456-69

Guyer, B. P. & Sabatino, D. (1989), The effectiveness of multi-sensory alphabetic phonetic approach with college students who are learning disabled, Journal of Learning Disabilities, 23: 43-4

Miles, M., Martin, D. & Owen, J. (1998) 'A Pilot Study into the Effects of Using Voice Dictation Software with Secondary Dyslexics Pupils', Devon, Devon Education Authority (Unpublished)

Miles, M. (2000) A Pilot Study into the Effects of Using Voice Dictation Software with Primary Dyslexic Pupils, Devon County Council (Unpublished)

Oakland, T., Black, J. l., Stanford, G., Nussbaum, N. L. & Balise, R. (1998) An Evaluation of the dyslexia training Profram: A Multisesnory method for promoting reading in students with reading disabilities, Journal of Learning Disabilities, 31:140-47

Raskind, M. H. & Higgings, E. L. (1999) Speaking to Read: The Effects of Speech Recognition Technology on the Reading and Spelling Performance of Children with Learning Disabilities, Annals of Dyslexia, Vol 49, pp251 - 281

Sanderson, A.H. (1999). 'Voice Recognition Software: A Panacea for Dyslexic Learners or a Frustrating Hindrance ?', Dyslexia, Vol 5 No 2

Appendix 1

Tests used for Dutch Participants/Control Groups were as follows:-

• Reading:-
  Brus (one minute reading test)
  Klepel (two minute nonsense word reading)
  AVI test :reading texts aloud
  Silent reading task
• Spelling:- PI spelling test, B version
• Free writing:- 10 minutes on 'What would you do with a million Gilders'
• Informal auditory memory (unrelated, single word recall)

Tests for English Participants/Control Groups were as follows:-

• Reading:-
  One minute reading test
  PhAB (nonsense words)
  Oral reading test (miscue)
  Silent reading task/comprehension
• Spelling:- WRAT 3 to give a spelling age equivalent
• Free writing:- 10 minutes on 'What would you do with a million Pounds'
• Informal auditory memory (unrelated, single word recall)