Spellchecking software –some initial findings

I’ve been developing a minor obsession with spell-checking software lately. This arose from noting that the spell-checking in Google Drive (previously called Google Docs) was pretty good and seemed to give decent results.  These results were also reached without the need for two separate checks – spelling and homophones.  The reason for these results and dispensing with the homophone check is that Google Drive analyses the spelling errors in context of the sentence. The more commonly used spellcheckers seem to analyse each error individually.  The other power that Google Drive has is a database of all the spelling errors made by everyone who has ever typed in a Google search.

What I wanted to do was compare the outcomes of Google Drive and some other, newer spell checkers with the more commonly recommended packages.  I also wanted to test the spell checkers on text produced by individuals with literacy difficulties (not necessarily diagnosed dyslexia).  This was primarily to see whether the contextual checking had an impact but also to know whether as an Assistive Technologist and Assessor I should be recommending a different approach for learners.

I hope that this builds upon the previous spellchecker research by Abi James and E.A. Draffan, and CALL Scotland.

Disclaimer

These are initial findings and are based on the following:

  1. I am only looking at the spellchecking element of the software being tested.  I am aware that the other features may be of benefit but they are hard to quantify.
  2. I am aware this is a very limited study.  I hope to gather more data on which to test the software soon.
  3. I am not an experienced academic researcher and am happy to be challenged on any of my methodology or results.

Method

13 volunteers provided 2 typed paragraphs for the study. The first was free writing from a choice of subjects: My typical weekend, The best holiday I’ve ever been on or My favourite place and why.  The second paragraph was dictated to the volunteer and was constructed from commonly mis-spelled English words (paragraph is shown in notes below).

The freely typed paragraphs ranged from nothing (a common occurrence when presenting a blank page to someone with literacy difficulties) to eight sentences.

The paragraphs were then tested with each spellchecking package and each error correction was scored 1,2,3,4 or 5.  If the correct spelling was first or only in a list it was scored 5, second in the list was scored 4 etc. If the error was not found, the correct spelling was not given or was below fifth position in a list it was scored 0.

The spellchecking software was used “out of the box”.  The only change made was that any “learning” functions were turned off so that each test did not influence subsequent ones.

Results

An average (mean) was taken for all the results from each software package and these can be seen in the graphs below. (Click on a graph for a larger image).

Bar graph showing software titles on horizontal axis and score out of 5 on vertical axis.  Scores are:  Verity Spell 4.18, Say So 4.10, Ginger 3.86, Ghotit 3.85, read and Write Gold 3.67, ClaroRead 3.55, Word 3.48, GoogleDrive 3.46

Bar graph showing software titles on horizontal axis and score out of 5 on vertical axis.  Scores are:  Ghotit 3.74, Ginger 3.66, Claro 3.24, Google Drive 3.16, Verity Spell 3.15, Word 3.04, Say So 2.85, Read and Write Gold 2.42

I’ll leave the graphs to speak for themselves, but will add the following points:

  1. To the best of my knowledge the software titles which check contextually are: Ghotit, Ginger and Google Drive.
  2. It is possible that the contextual spell checkers are more successful on the dictated paragraph as the sentence construction will be typical English. This is not the case for all the free writing.
  3.  It is hard to say which is the better subject for testing: free writing or dictation. Free writing enables participants to avoid tricky words, but is likely to be closer to their writing style. Dictation probably forces more errors, but give the contextual checkers an unfair advantage as the sentence structure is more typical English.

Notes

The dictated paragraph:

On Wednesday my friend told me that he would be moving to Edinburgh in February.  At first I did not believe him because lately he had been acting a little weird.  I definitely didn’t want to embarrass him though so I asked whether he had sorted out any accommodation.  He said there were a lot of beautiful flats particularly in the city centre.  He definitely thought it was unnecessary to live right in the centre, about two or three miles out would be acceptable.

Fil McIntyre, Assistive Technologist, The BRITE Initiative

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s