Proofreading Academic Writing: Human vs. Machine

Between designing and performing studies, applying for funding, attending conferences, writing manuscripts, and numerous other commitments, researchers may have little time available for proofreading. One rapid approach to this problem is to use an automated proofreader, which typically consists of a web-based service or downloadable plug-in that checks spelling, grammar, and style. These proofreaders offer several advantages, including being available around the clock, providing immediate results, and often being relatively inexpensive. These services may also provide educational explanations of errors and potential corrections. However, reviews elsewhere (such as here) have found that these programs miss certain errors and inconsistencies and suggest unnecessary changes, which may be a particular problem for authors writing in a non-native language. A digital proofreader will also overlook most gaps in logic due to a lack of understanding of the content. However, one could argue that human proofreading is also subject to oversights and misinterpretation.

Human editors vs. the automated tool Grammarly

Nevertheless, there is a little-explored area in which a human editor may be preferable over a computerized one, with particular relevance to academic research: assessing the use of field-specific terms. To test this hypothesis, we compiled a list of correct and incorrect sentences (see lists below) based on several commonly confused and frequently misused terms detailed in our clinical writing resource. For good measure, we added one correctly used unique term that may appear incorrect in nonclinical contexts: "past history." We then submitted this text for review by the popular proofreading service Grammarly, a “writing enhancement app” that performs checks for contextual spelling (including commonly confused words), grammar, punctuation, and style and word choice (including vocabulary use).

The results were clear: overall, although the automated proofreader did not mark the correct sentences as incorrect (with one exception, detailed below), it did yield false negatives; that is, it did not detect the errors in the incorrect sentences. In other words, the proofreading program indicated that the vast majority of both correct and incorrect sentences were correct. In particular, using the “General” setting, the system questioned the use of pronouns and passive voice, which are typically acceptable in clinical case reports, and noted that "past history" is redundant, even though this term is conventional in clinical writing (with over 500,000 Google Scholar hits). The “Academic” setting produced the same results. Finally, the “Technical” setting, though not sensitive to the pronoun and passive voice use, still flagged the term "past history." Microsoft Word's spelling and grammar check was similar, noting the passive voice and the apparent wordiness of "past history" but none of the errors in the incorrect sentences.

Notably, whereas the Grammarly website mentions that the tool can be used to “Grammar check medical reports,” the chief executive has stated that Grammarly “isn't intended for professional writers to achieve the next level of language mastery.” Based on the small study presented here, automated editing services such as Grammarly may be acceptable for basic proofreading but may not be as good a fit for more technical writing; this is where live subject-area experts have the advantage.

When we submitted examples of the correct (left column) and incorrect (right column) use of clinical terminology and phrasing to Grammarly, the automated proofreader, did not identify any of the errors in the use of the commonly confused or frequently misused terms:

Commonly Confused Terms

Correct

• A case of Parkinson's disease is presented.
• A 325 mg dose of aspirin was administered every 4 hours.
• The survival period was 5 years.

Incorrect

• A case of Parkinson's disease visited our clinic.
• A 325 mg dosage of aspirin was administered every 4 hours.
• The survival period was 80%.

Frequently Misused Terms

Correct

• Emphysema was correlated with shortness of breath.
• The patient was diagnosed as having asthma.
• Bacterial infection is an indication for antibiotic use.
• The patient was operated on in the hematology unit.
• Iron levels were a predictor of anemia.
• Smoking is a risk factor for lung cancer.

Incorrect

• Emphysema was correlated to shortness of breath.
• The patient was diagnosed as asthma.
• Bacterial infection is an indication of antibiotic use.
• The patient was operated in the hematology unit.
• Iron levels were a predictor for anemia.
• Smoking is a risk factor of lung cancer.

Unique Term

Correct

• The patient had a past history of hypertension and a current history of heart disease.