At the end of the 1990s, a machine did an extraordinary thing: Deep Blue, the chess-playing computer developed by IBM beat the world chess champion, Garry Kasparov, in a chess match. At that time, Dr. Piet Hut, an astrophysicist at the Institute for Advanced Study in Princeton said, It may be a hundred years before a computer beats humans at Go—maybe even longer.
At the end of 2016 and less than 20 years later—far fewer than Hut's predictions—a popular game server in Asia watched one of its Go players named "Master" dominate most of the world champions who played online. Players, desperate to find out who the new champion was, finally discovered within the next few weeks that "Master" was DeepMind's AI AlphaGo. By May of 2017, "Master" had defeated the Go world champion and by October of the same year, Google announced it had a more sophisticated version of AlphaGo.
So, what does this have to do with proofreading? Well, a lot—especially since there is a lot of talk now in business and academic circles about the advantages of machine-based proofreaders over human proofreaders. For many, it's a cost and convenience factor, especially with services like Grammarly® offering free proofreading available through easy-to-use apps and cloud-based software.
If computers can be taught to beat the human world champions of Go and Chess, surely they can be taught to proofread better than humans too, right?
Editing is no game
The answer is no, according to Grammarist.com, an online resource for all things grammar. Since the popular proofreading software, Grammarly, advertises that its algorithms flag context-specific corrections for grammar, spelling, wordiness, style, punctuation, and plagiarism, the authors decided to test it out. They opened a Grammarly Premium account and then ran a series of sentences through it to check its ability to catch common errors. They then used a few sentences from an article published on a lesser-known website to see how well Grammarly's plagiarism checker works, including rewording them slightly to see what level of rewording was needed to not be caught by the checker as plagiarism.
According to the authors, In the tests that were quantifiable, Grammarly was asked to check for forty-three mistakes, and it managed to find thirty-one of them. That's 72 percent. But the numbers only tell part of the story.
In an article on the same topic, Forbes contributor Ben Kepes agrees with that assessment. Grammarly isn't a complete replacement for manual proofreading,
he writes. I was surprised at the number of errors it made—both false positives and not picking up on obvious mistakes in both spelling and grammar. The number of times I had to change manually 'fo' to 'of' was a little annoying. I'm not sure I'd be quite as brutal as one review I read, Grammarly does, in my mind, have a part to play in spelling and grammar checks. But for someone looking for a single solution, Grammarly probably isn't it.
So, what do the machines miss?
In the few quantitative tests reviewers have discussed online, Grammarly, Microsoft Word, and other spellcheck or proofreading programs tend to miss certain types of mistakes. Of these mistakes, academic writers seem to have the most to lose in relying solely on a computer-based proofread of their writing.
For example, most software and cloud-based proofreading programs will flag writing that is in passive voice (e.g., "The presentation was completed by our group" [passive voice] vs. "Our group completed the presentation" [active voice].) Passive voice is commonly used in higher-level academic research and having a program flag each use of it is time-consuming when the writer has no intention to change it. Other problems the reviewers ran into was the use of terms like "past history," which is commonly used in the fields of health and psychology but flagged by the software as being incorrect.
Misused words
Another issue missed by proofreading software is a word that has been misused in context. Since context must be carefully considered to determine which word is best to use in a particular instance, proofreading programs have a difficult time catching these mistakes. This can happen with the incorrect use of some of the following words:
- affect vs. effect
- lead vs. led
- historic vs. historical
- insuring vs. ensuring
- lay vs. lie
- bring vs. take
- nauseous vs. nauseated
- comprise vs. compose
- fewer vs. less
- farther vs. further
Repetitive sentence structure
Writing with repetitive sentence structure is a common problem, especially for writers who are ESL learners—and it's one problem a computer program won't recognize. Take the following example:
This is bad writing. There are no grammar mistakes. Yet it is still bad writing. Can you guess why? Read it aloud. You'll figure it out then.
Repetitive sentence structure can be easily fixed by changing up the length of sentences and including more dependent clauses. However, it will take a human proofreader and editor to point this out to you and to help take your writing to a more academic or professional level.
Tense consistency
As a professional editor, one of the most common problems I've seen from over 10 years of editing for clients is tense consistency. A writer might start a paragraph in present tense but suddenly switch tenses mid-paragraph, causing confusion for the reader. Or worse, a writer might start a sentence, then suddenly switch tenses mid-sentence. These are issues that proofreading software will rarely catch.
Take the following example:
If the hospital wanted to have fewer patients, it will need to change its admission policies.
In the above sentence, the writer first used a past tense verb (wanted), then suddenly switched to a future tense verb (will need). This is grammatically incorrect and is often seen in poor writing.
Articles or words completely missing
If you happened to leave out a word in your essay, especially if that word is an article (the, a, an), it is highly unlikely that proofreading software will catch it. Also, if you happened to use "an" instead of "and", don't trust a software program to catch it.
Case study #1 – Online content
The only way to compare human editors to machines using algorithms is to pit the two against each other in multiple documents to see which does a better job across genres and writing styles. To do this, we used an original document submitted to professional editors on ServiceScape.com.
View the original online content and the human editor's corrections.
We then purchased a premium membership for Grammarly.com, which is advertising a current rate of $29.95 per month for its services. For the first comparison, we used a piece of web content that would be found on a company's website or marketing material. The human editor found multiple changes in the first and second paragraphs that were completely missed by Grammarly. Specifically, the human editor:
- Revised the first sentence to make it stronger and in more active voice (changing "developing" into "that develops").
- Changed a preposition (from "to" to "for"), making the sentence more logical.
In the third paragraph, both the human editor and Grammarly caught the misplaced punctuation (a comma that should be inside of the quotation marks instead of outside of it). However, that's the only similarity in the editing styles.
The human editor changed "spread" to "distribute" (which is more professionally written) and shorted the first sentence to give it more of a hook at the beginning. Obviously, a machine can't think this deeply about word choice and sentence structure for effect, but the algorithm did suggest changing "unparalleled" to "unique," which seems like an odd (and even unnecessary) change.
Under "Our services include," Grammarly made a good suggestion to change "key" to "vital" (since "key" is often overused in business writing) but then made an error by suggesting that I change "succeed" to "success," which changes the original intent of the writer. It then suggested I change "distinguished" to "unique" (again!) and it seems as if Grammarly really likes the word "unique", although it shouldn't have suggested it twice in one paragraph. Meanwhile, the human editor caught an illogical sentence ("Tailoring your voice aligned with your audience") and changed it to read coherently and in active voice.
In the sixth paragraph, the human editor caught three distinct instances in which the sentence was wordy and could say the same thing with fewer extraneous words. Apparently, this isn't a concept Grammarly's algorithms can understand.
Finally, both the human editor and Grammarly caught an error with a missing comma. The human editor, however, found that it sounded better to "share stories with the press" than to "share stories in the press." The human editor also found that a sentence was written illogically ("We also help boosting media coverage to our clients") and changed it to the more logical ("We also help boost media coverage for our clients"). Grammarly did not make any of these suggestions.
Results
Although Grammarly caught several important errors, it was unable to suggest changes in wording to make the content more concise and logical. Particularly, it doesn't seem to be able to catch when wordiness is happening, which can make web content less readable/sharable. For these reasons, the human editor wins this round!
Case study #2 – Academic writing
Now let's look at academic writing.
View the original academic document and the human editor's corrections.
Let's begin by looking at the Abstract. Abstracts in academic writing must be briefly and clearly expressed, which is why the human editor changed the client's opening sentence.
Original sentence: A formulational debate is a debate over whether certain definitions of scientific realism and antirealism are useful or not.
Edited sentence: A formulational debate is a debate over the usefulness of certain definitions of scientific realism and antirealism.
After running the original document through Grammarly, the program did not make any changes for better readability and succinctness in the abstract. It did, however, point out that "formulational" isn't in its dictionary and should be added if it is correct. It also pointed out that formal, academic writing should not use "I", which the human editor failed to note. However, this rule isn't consistent across all fields and many fields do use personal pronouns in academic writing or publication, especially when research was conducted.
In the section beneath the header "Formulational and Epistemological Debates, and the No-Miracles Argument," the human editor made several comments regarding the logic of the argument and missing elements of a list. Obviously, Grammarly does not have the capacity to do this type of in-depth editing for academic work. The program only noted the use of "I" multiple times and possible incorrect spelling of "formulational".
Further into this section, the human editor made changes to a sentence to reduce wordiness and confusion (for example, the original sentence contained "or not" at the end, which could cause confusion for the reader).
Original sentence: Under Putnam's formulations, realists and antirealists also have been in formulational debates over whether certain formulations can overcome the pessimistic induction or not.
Edited sentence: Putnam's formulations have also engendered formulational debates over whether certain formulations can overcome the pessimistic induction.
Grammarly was unable to make this sort of change, but it did suggest "successful present" be changed to "present successful"—a change which, if made, would ruin the parallel structure of "successful past," which is mentioned later in the sentence. While not a grammar error, this is a suggestion that would make the writing weaker instead of better.
In section 3.1, the human editor made several changes to the following sentence: "Teleological explanations were regarded as legitimate in ancient science, but only mechanical explanations are regarded as legitimate in modern science." Specifically, the editor took the sentence out of passive voice and made it active and changed "but" to "whereas" to make the writing more academic and logical. This section also used "such" without using "as" in referring to specific ancient philosophers ("…, such Aristotle and Ptolemy"). Both Grammarly and the human editor caught this and suggested correct changes.
The remaining parts of the document show extensive comments made by the human editor concerning logical inconsistencies and counterintuitive reasoning in the author's points. Obviously, Grammarly is unable to take this comprehensive approach when editing or look at the logic of the argument in a way that a human can.
Results
In this particular case study of academic writing, the clear winner was the human editor. This is primarily the case because he was familiar with the subject and could address the author's reasoning as well as his/her grammar.
Case study #3 – College entrance essay
A college entrance essay is your one chance to stand out above other applicants, while expressing your personal and academic goals. However, for this particular case study, the applicant was lucky that he or she chose a human editor over Grammarly.
View the original college entrance essay and the human editor's corrections.
The first paragraph shows why the applicant was lucky. Here is what the author originally wrote as in introduction:
I want to become an equity analyst in private equity or fund in Singapore after MBA programme. My previous education background and working experience in accounting and auditing build me solid accounting techniques and financial data analysis skills which are demanded for an equity analyst. In long term, after gaining enough investment experiences, I would like to manage a small fund for my family.
At first glance, there are several major errors that stand out, all of which were caught by the human editor. For example, "in private equity or fund" is illogical; an article or pronoun is missing from "after MBA programme"; "build me solid" is illogical; and "In long term" needs "the" added. Grammarly did not catch all of these errors, which was surprising since they were so obvious, although it did suggest that "in long term" be changed.
It doesn't get better in the following paragraph. Grammarly did catch "can help me to acquire those knowledge" and suggested the author replace "those knowledge" with "that knowledge," but did not catch the missing pronoun "I" in "Although majored in accounting and finance."
In the final paragraph, Grammarly caught multiple errors, but did not make the same changes suggested by the human editor to help the writing flow more naturally.
Results
For something as important as college admissions writing, hiring a human editor is the smartest move—especially if the author's first language is not English. Grammarly missed glaring and obvious mistakes, which can mean the difference between being accepted or denied into an institution of higher learning.
Case study #4 – Cover letter
For our fourth comparison, we looked at a cover letter submitted to both Grammarly and a human editor.
View the original cover letter and the human editor's corrections.
Grammarly caught several errors, including suggesting replacing "fast-faced" with "fast-paced".
Of course, the human editor caught these same errors. This particular case study is similar to the academic writing case study in that the human editor offered helpful advice on top of editing services. For example, the editor suggested that the writer avoid contractions in a professional letter. The editor also suggested that the writer avoid including details of her three boys under 5 years old, commenting that the employer might take this as a negative trait in an employee by assuming that the employee will be absent from work a lot to take care of her three young children.
Beyond that, neither the human editor nor Grammarly caught the problem with hyphens in "five-years-old" (it should not have hyphens between the words unless it is used as an adjective). However, the human editor did suggest removing that phrase entirely, so that could have been the reason she avoided pointing out the error.
Results
Grammarly did a better job of catching errors in this round, so from a simple proofreading standpoint, it was a tie. However, the human editor included important advice concerning the need to remove personal information in a cover letter for a job search, which means the human editor still won this round.
Case study #5 – Technical writing
For this comparison, we used technical instructions written for a new App that manages a person's contacts.
View the original technical instructions document and the human editor's corrections.
Grammarly caught several instances in which a sentence ended with a preposition and correctly suggested that passive voice should not be used so much.
However, the human editor once again proved to be more helpful for this particular client. For example, the human editor removed an entire sentence about how the App keeps your family and friends from having to post their addresses and contact information on social media (which most people know not to do). The human editor also rewrote some long sentences into shorter, easier-to-read instructions.
One strange suggestion Grammarly gave was to remove "personal" when referring to one's "personal link" because it "creates tautology." While this is a complicated way to say there might be repetitive information, it isn't an example of tautology because the instructions related to a "personal link" needed the word "personal" in it to clarify which link should be used.
Results
As with other case studies, the results of this round leave the human editor as the clear winner. Grammarly's algorithm is limited in its ability to make sentences more concise, and if rewriting is needed, Grammarly does not offer suggested rewrites.