Published in the PATOSS bulletin in 2004, the article that examines the accuracy of spell checkers in software packages for supporting dyslexic learners and portable spell checking devices.

This article was written during August 2004 for the PATOSS bulletin by Abi James & EA Draffan, revised with updated product links February 2017.

For many years it has been accepted that word processors and spell checkers help overcome the many spelling difficulties encountered by dyslexic children and adults. However, studies such as McArthur et al (1996), Nisbet et al (1999), Montgomery et al (2001) and Pedler (2001) found that spell checkers’ ability to correct misspellings varied greatly and most could not correct a significant proportion of the errors. Factors affecting the spell checkers’ accuracy included the severity of the error and the difficulty of the spelling. In the past 10 years a number of spell checkers have been developed specifically for dyslexic learners. These spell checkers tend to be known as “phonetic” spell checkers because they look for errors based on letter-sound combinations. However a pilot study in April 2003 found that current spell checkers were still unable to correct 30% of misspellings (TechDis, 2003). This type of study, highlighting the important differences between spell checkers, can help teachers, parents, needs assessors and purchase decision makers identify the best spell checker for an individual.

Over the past year we have undertaken an investigation of how well current spell checkers cope with the types of spelling errors made by learners with dyslexia. In doing so we have developed a methodology for testing and ranking spell checkers based on a list of 264 spelling errors from dyslexic children and adults, classified by:

  • how complicated the word was to spell. Each spelling was categorised as Key Stage 2 (children aged 7-11), Key Stage 3 (children aged 12-14) or Key Stage 4 (children aged 14-16) depending on when a learner would be expected to learn it.
  • how severe the error was using a proportion of correct letter pairs in the letter, known as bi-grams (Montgomery et al, 2001).
  • how the spelling had been attempted using a phonetic or visual strategy (based on the number and placement of ascenders and decenders). Some errors were classified as bizarre as there was no clear attempt to spell the word phonetically or visually.

Why do spell checkers fail to suggest the correct word?

There can be two reasons why spell checkers fail. The first is experienced by all users. That is that the spell checker does not have the correct word in its dictionary. This happens frequently with proper nouns such as names or when checking specialist terminology. Most spell checkers have an ‘add to dictionary’ function which reduces this problem over time but can lead to incorrect spellings being added to the database by dyslexic learners. This leads to spellings being double checked using specialist dictionaries or resources. Subject-specific spell checker add-ons, such as those developed by Spellex Handheld Medical Spell Checker (Spellex) or programs such as Clicker (Crick) can help reduce this issue.

Spellcheckers which are more accurate and now available since this article was written are available:

Software packages:

Texthelp Read & Write – a writing support toolbar with speech feedback and a phonetic spell checker. [New version now available – Texthelp Read & Write]

There are a range of ClaroRead products from Claro to support people with dyslexia and who require text-to-speech feedback with spelling options

Portable spell checkers from Franklin

[New version now available Franklin Pocket Collins Dictionary/Thesaurus, DMQ-570] [New version now available Franklin Speaking Dictionary/Thesaurus, DMQ-2110]

The second reason why a spell checker may fail to suggest the correct word is when the spell checker is unable to match the attempted spelling to the right word; usually caused by a combination of the attempted spelling being too far from the correct word and the limits on which the spell checker searches through its dictionary (Mitton, 1996).

Studies by Damerau (1964) and Pollack & Zamora (1984) found that between 80% and 95% of spelling errors could be grouped into 4 simple categories:

  • one letter wrong (peaple)
  • one letter omitted (peple)
  • one letter inserted (peopple)
  • two adjacent letters transposed (pepole)

and that the majority of spelling errors contained a correct first letter (Yannakoudakis & Fawthrop 1983) . We collated a list of 264 errors, predominantly from dyslexic adults. Only 31% fell into these simple, typing error categories and 7% had the first letter incorrect. This list contained spelling errors of words with a wide range of difficulty from regular spellings, such as “but” and “cook”, to more complicated, irregular spellings such as “enthusiasm” and “jeopardise”. Although this list is not meant to be representative of how dyslexic children and adults spell (it cannot be used to say that spelling errors made by a dyslexic child or adult are different to those of a non-dyslexic), it does show that dyslexic learners consistently make spelling errors outside the rules used by software developers to develop spell checkers.

What makes a spell checker effective for a dyslexic learner?

Many dyslexic learners can find spell checkers difficult to use, even if the correct spelling is suggested. McArthur et al (1996) found that dyslexic learners missed 27% of correct suggestions in his study of spell checking programs. The number of suggestions missed, despite being correct, increased if the word appeared further down the suggestion list. If the correct spelling appeared first in the list then 84% of the words were correctly chosen but if it appeared further down the list the figure fell to 72%. To understand why this might happen we must examine the processes and steps involved in identifying and selecting the correct word from the suggestion list. If a dyslexic learner is presented with a list of suggestions they must:

  1. 1.    Decode and comprehend each suggestion. This may be difficult if they have poor phonological processing and reading skills.
  2. 2.    Compare the word they have written to the correct word as they remember it. This task relies on keeping the suggestion and the word the learner is attempting to spell in their working memory. They must also recall the original word quickly and accurately. If the learner relies on visual memory skills then a similar looking suggestion may be chosen in error.
  3. 3.    Compare the most likely suggestion to all the others the spell checker is making to ensure it is the right one. Again this task relies heavily on working memory as the learner is required to remember a number of suggestions at any one time. The learner must also remember to read the whole list, including those not shown on the first screen. If the learner has poor attention they may easily forget that more suggestions may be available by scrolling down.

Therefore to use a spell checker effectively good visual and phonological processing, working memory and attention skills are required, all areas at which dyslexic learners may struggle. How well they use the spell checker will depend on their skill set. If they have poor working memory and phonological decoding skills but good visual processing they will rely on visually scanning the suggestion list to identify the right word. If the suggestion list is split over a number of screens and the user has to scroll through them, they will be relying on their visual memory to remember and compare the suggestions. Therefore having the correct spelling always presented on the first screen of the suggestion list will greatly improve the efficiency of the spell checker for these types of users. For those users who have poor working memory and visual processing skills, they must rely solely on phonological decoding skills to read each suggestion in turn. In this case it is vital that the correct spelling appears as close as possible to the top of the suggestion list. McArthur et al (1996) and Montgomery et al (2001) concluded that the correct spelling must appear in the top 3 of the suggestion list for the spell checker to be effective. Therefore to measure the effectiveness of a spell checker we must look not only at the number of words where the correct spelling is suggested but also at the number of correct spellings that are ranked in the top 3 of the suggestion list and the number that are ranked on the first screen.

Leary (2002) examined the effect of teaching children with Specific Learning Difficulties how to use a spell checker effectively through editing and re-checking errors. She reported a significant improvement in the spelling abilities in 10 of the 11 children included in the study. Notably, Leary also concluded that the severity of the error was not the main factor as to whether the spell checker helped with spelling; it was more dependent on the modalities used to spell. Children who depended on a combination of auditory and visual skills or solely on auditory skills tended to attempt spellings phonetically. The spell checker was more capable of correcting these types of errors than those based on a visual representation of the word as made by those children who relied solely on visual skills. Therefore it is important to examine how spell checkers vary with different strategies used to spell a word.

Testing Methodology

For accurate spelling correction the spell checker needs to (in order of importance):

  1. 1.    Identify an error as incorrectly spelt.
  2. 2.    Suggest the correct word.
  3. 3.    Suggest the correct word at the top of a suggestion list.

The tests undertaken only included misspellings, with real words and homophones removed. It was assumed the spell checkers would identify these words as erroneous. To measure (2) and (3) each error was individually checked through each spell checker. Where the spell checker suggested the correct word the position in the suggestion list was noted and if it failed this was also taken into account. The software packages were tested on a fresh installation with new user files and auto-correct functions turned off. No alterations were made to the settings so as to test the spell checkers performance ‘out of the box’.

The results of 4 software applications and 3 handheld spell checkers available in the UK in 2004 are presented here.

Table 1 presents the percentage of correct suggestions made for each spell checker along with the percentage that were ranked in the top 3 and appeared on the first screen. The highest results at have been highlighted. Overall the best performing spell checkers, Write:Outloud and Spell Catcher (highlighted on the table), managed to correct 71% of our errors with the majority of the spell checkers correcting at least 65%. This is comparable to results found by McArthur et al (1996), Montgomery (2001) and Pedler (2001). However, one spell checker, the Franklin DMQ 450, was only able to correct 52% of the errors. This accounted for an additional 51 errors not being corrected if the worst performing spell checker was used.

Overall resultsWord 2003Write: Outloud v3Read & Write 7Spell Catcher Plus v2Franklin LWB216Franklin CollegiateFranklin DMQ450
% corrected66.67%71.21%65.53%71.21%70.45%65.15%51.89%
% corrected in top 359.09%54.55%52.27%53.41%62.88%54.55%47.73%
% corrected on 1st screen65.91%62.12%63.64%67.80%58.33%64.39%47.73%

Table 1: Overall percentage of corrected by each spell checker

Each correct spelling had been classified by its difficulty prior to testing. These categories were based on which stage in the English National Curriculum the spelling would be expected to be learnt and can be thought of as regular spellings taught at Key Stage 2 (KS2), irregular spellings taught at Key Stage 3 (KS3) and less common words required by Key Stage 4 (KS4). The results are shown in Table 2.

Key StageWord 2003Write: Outloud v3Read & Write 7Spell Catcher Plus v2Franklin LWB216Franklin CollegiateFranklin DMQ 450
2% corrected73.56%80.46%79.31%79.31%82.76%66.67%47.13%
% corrected in top 362.07%58.62%56.32%56.32%70.11%56.32%40.23%
% corrected on 1st screen73.56%71.26%75.86%75.86%63.22%65.52%40.23%
3% corrected63.21%60.38%56.60%62.26%66.04%63.21%50.94%
% corrected in top 358.49%42.45%43.40%43.40%59.43%50.00%48.11%
% corrected on 1st screen61.32%50.00%54.72%59.43%53.77%62.26%48.11%
4% corrected63.38%76.06%61.97%74.65%61.97%64.79%59.15%
% corrected in top 356.34%67.61%60.56%64.79%59.15%59.15%56.34%
% corrected on 1st screen63.38%69.01%61.97%70.42%59.15%64.79%56.34%

Table 2: Percentage of words corrected, by difficultly of the spelling.

The spelling errors were also categorised by how severe the error was and whether a visual and/or a phonetic strategy had been used to spell the word. These results are displayed in Table 3 and 4 respectively.

Severity of ErrorWord 2003Write: Outloud v3Read & Write 7Spell Catcher Plus v2Franklin LWB216Franklin CollegiateFranklin DMQ 450
Simple errors% corrected94.74%90.79%86.84%92.11%86.84%89.47%75.00%
% corrected in top 394.74%84.21%78.95%85.53%82.89%77.63%69.74%
% corrected on 1st screen94.74%89.47%86.84%92.11%81.58%88.16%69.74%
Moderate errors% corrected62.25%69.54%60.93%68.87%70.20%59.60%47.02%
% corrected in top 351.66%49.01%45.03%47.02%60.26%49.67%43.05%
% corrected on 1st screen60.93%57.62%57.62%65.56%54.30%58.94%43.05%
Severe errors% corrected27.03%37.84%40.54%37.84%37.84%37.84%24.32%
% corrected in top 316.22%16.22%27.03%13.51%32.43%27.03%21.62%
% corrected on 1st screen27.03%24.32%40.54%27.03%27.03%37.84%21.62%

Table 3: Percentage of words corrected, by severity of the error.

Strategy used to spell wordWord 2003Write: Outloud v3Read & Write 7Spell Catcher Plus v2FranklinLWB


FranklinCollegiateFranklin DMQ 450
Phonetic% corrected83.93%86.61%79.46%86.61%85.71%81.25%71.43%
& Visual% corrected in top 378.57%73.21%66.07%74.11%76.79%67.86%65.18%
% corrected on 1st screen83.04%79.46%78.57%84.82%72.32%80.36%65.18%
Phonetic% corrected68.06%70.83%68.06%70.83%68.06%69.44%51.39%
% corrected in top 358.33%55.56%55.56%51.39%61.11%54.17%48.61%
% corrected on 1st screen68.06%62.50%66.67%66.67%54.17%68.06%48.61%
Visual% corrected48.44%56.25%53.13%56.25%62.50%45.31%31.25%
% corrected in top 337.50%32.81%35.94%31.25%54.69%43.75%28.13%
% corrected on 1st screen46.88%42.19%48.44%50.00%51.56%45.31%28.13%
Bizarre% corrected12.50%25.00%6.25%25.00%6.25%12.50%0.00%
% corrected in top 312.50%6.25%6.25%6.25%6.25%6.25%0.00%
% corrected on 1st screen12.50%18.75%6.25%25.00%6.25%12.50%0.00%

Table 4: Percentage of words corrected, by strategy used to spell the word.


The Franklin Literacy Word Bank (LWB 216) was found to be the best spell checker for suggesting the correct spelling in the top 3 and also did very well overall but as it only displays 2 suggestions at once it is not as useful for learners who rely on visual comparison of suggestions to identify the correct word. Spell Catcher Plus provided the highest percentage of correct suggestions on the first screen as it displays 10 suggestions at a time.

The Franklin DMQ 450 failed to correct 48% of the spelling errors which makes us unable to recommend it for dyslexic learners. This handheld spell checker has recently been released, replacing the Franklin DMQ 440N, which had a smaller dictionary. However, this older model was a better spell checker when checked with our list of errors. It is important to remember that a new version or a larger dictionary may not necessarily lead to improvements in the performance of a spell checker.

All the spell checkers bar the Franklin DMQ450 coped best with the regular spellings in the KS2 category and least well with those in the KS3 category. Although the KS3 category contained more words than the other categories we found no relationship between the severity of the error and our categorisation by difficulty of word. This seems to indicate the spell checkers in general find it more difficult to correct attempts at irregularly spelt words. This could be because they have been designed to search based on regular spelling rules.

The Franklin Literacy Word Bank managed to correct the most errors in both the KS2 and KS3 category. This handheld spell checker is based on the Oxford Primary dictionary and was designed for primary aged children. However our list of errors show that dyslexic learners of all ages make mistakes with frequently used words and those considered easy to spell. So spell checking products aimed at a younger age range can be considered useful for supporting all ages of dyslexic learners.

Word 2003 was the only non-specialist software package included in the tests. Overall it coped very well, correcting 74% of the spelling errors. However, when we look at the breakdown of results by severity of the spelling error (Table 3) it can clearly cope with simple spelling errors very well (correcting 94%) but it struggles with the more severe errors, as we would expect from a program optimised for typing errors as much as spelling errors.

At the KS4 level two software packages stood out from the rest Write:Outloud and Spell Catcher  correcting 75% or more of the errors. Although very different programs functionality wise, they both use the same Franklin dictionary in their spell checking functions.

It was clear from our list of spelling errors that many dyslexic learners do not spell words solely as they sound or phonetically as frequently stated. 30% of errors could not be matched to the correct spelling by phonetics. However the spell checkers that have been developed specifically to cope with phonetic errors – Write:Outloud, Read & Write, the Franklin Literacy Word Bank and Spell Catcher Plus – all managed to correct the majority of visual errors. As learners who use a visual spelling strategy are also more likely to recognise the correct spelling by its shape it is particularly important that the correct spelling appears on the first screen for visual strategy errors. The Franklin Literacy Word Bank only displays 2 suggestions at a time and so is more taxing on memory and decoding skills than spell checkers that present a longer list for these visual spellers. As Read & Write and Spell Catcher both present 10 suggestions at a time, they are more effective for these users.

Read & Write showed the least variation in results across the severity of error and the strategy used to spell the error. It also corrected the highest percentage of severe errors. This indicates that this checker is a good solution in an environment where there will be a wide range of users with different spelling abilities and strategies. As Read & Write’s spell checker can also be adapted to each user the results should improve with time for individual users.


This study has enabled us to develop a method to quantify the abilities of spell checkers to handle spelling errors. Given that spell checkers change so frequently (all bar 3 of the spell checkers discussed here have changed in the past 12 months) and that new developments do not always lead to improvements in the accuracy of the spell checker, quantifying their effectiveness needs to be repeatable. The method used here is repeatable and has currently been applied to 18 spell checkers, results from which will be published at This study has also highlighted that individual’s learning styles and skills affect the spell checkers’ effectiveness. It is important to match the strengths and weaknesses of the individual to the correct tool to support their spelling.

Dr Abi James is the Product Manager for Software and Assistive Technology at Iansyst Ltd while EA Draffan is an independent consultant on Assistive Technology. Both are also Visiting Research Fellows with the Assistive Technology Group at the University of Manchester .



Crick (2004). Wordbar v2. Crick Software Ltd.

Damerau, F. J. (1964). A technique for computer detection and correction of spelling errors. Communications of the A.C.M.,7 , pp171-176.

Learhy, M. (2002). Spelling, Spelling-Checkers and Dyslexia. CESI Conference January 2002 St.Patrick’s College, Dublin.

MacArthur, C. A., Graham, S., Haynes, J. B., & DeLaPaz, S. (1996). Spell checkers and students with learning disabilities: Performance comparisons and impact on spelling. Journal of Special Education, 30 , pp35-57.

Mitton, R. (1996) English Spelling and the Computer. Longman, London , 207p.

Montgomery , D.J., Karlan G.R. & Coutinho, M. (2001). The Effectiveness of Word Processor Spell Checker Programs to Produce Target Words for Misspellings Generated by Students With Learning Disabilities. Journal of Special Education Technology, 16.

Nisbet, P., Spooner, R., Arthur, E. & Whittaker, P. (1999). Supportive Writing Technology. Call Centre, Edinburgh , 223p.

Pedler, J. (2001) Computer Spellcheckers and Dyslexics – a Performance Study. The British Journal of Educational Technology, 32/1, pp. 23-38.

Pollock, J. & Zamora, A.(1984) Automatic spelling correction in scientific and scholarly text. Communications of the A.C.M.,27 , no.4, pp358-368.

Spellex (2003). Spellex Medical spell checker. Spellex Ltd

TechDis (2003). Which spellchecker is best?. TechDis Ezine, April 2003

Yannakoudakis, E.J. & Fawthrop, D (1983).The rules of spelling errors. Information Processing and Management,19 , no.2, pp87-99

Product references and links updated July 2006

Author: Abi James

Published: 12 Jul 2006