Recently, we’ve seen increased interest in language similarity reduction in manuscripts, fueled by the use of AI-based plagiarism checker software platforms that aim to detect similarities between students’ and researchers’ submitted work and previously published texts.

How do these tools work?

Two of the most commonly used software platforms for detecting similarity in manuscripts are Turnitin and Crossref/iThenticate.

Turnitin is more commonly used when assessing students’ coursework and theses. This software crosschecks manuscripts against a large online dataset of academic publications (e.g., articles, books, websites, theses) and flags any similarities identified as potential instances of plagiarism. Understanding types of plagiarism is crucial for both students and researchers to maintain academic integrity. (Source: https://www.apu.apus.edu/area-of-study/education/resources/how-does-turnitin-work/)

Crossref is more commonly used by researchers and academic publishers. Its similarity check service is based on the iThenticate tool, which was originally developed by Turnitin and works similarly to detect potential similarities between the submitted document and works in a large online dataset of publications. For those seeking alternatives, it’s worth noting how you might compare Turnitin and Grammarly for different use cases. (Source: https://www.crossref.org/services/similarity-check/)

What is the similarity score, and how is it calculated?

It may seem reasonable to assume that the similarity score indicates the percentage of a paper that has been plagiarized. However, this is an oversimplified and erroneous assumption. According to iThenticate’s user guidance regarding the Similarity Report produced by its interface：

“The similarity score is simply the percentage of text in a submission that matches other sources”

In other words, just because a paper receives a high similarity score does not mean that it contains plagiarism. It’s important to distinguish between similarity and plagiarism vs academic integrity violations.

“It is perfectly natural for a submission to match against some of our database,” according to iThenticate. The user guide recommends that authors and publishers use the report merely as a starting point during review and suggests that good judgment should be used to determine whether academic misconduct has actually occurred.

When submitted for a similarity check, a long document may receive a high score simply due to the large volume of text and the unavoidable repetition of common terminology and phrasing structures. Additionally, in-text citations and reference lists (which vary depending on citation styles comparison such as APA, MLA, or Chicago), which are directly replicated from the sources to documents, will automatically be flagged as “similar” during a cross-check against a literature database. It’s also worth noting the difference between copyright infringement vs plagiarism, as these are distinct legal and ethical issues. (Source: https://guides.ithenticate.com/hc/en-us/sections/22768794980749-The-Similarity-Report)

Should authors be concerned?

AsiaEdit often receives requests to assist authors whose work has received high similarity scores, as in the image below.

We’ve found that most of these scores are due to many small, arbitrary incidences, defined as those having a similarity score of 1% or less, rather than to large incidents of actual concern.

A large proportion of the instances flagged in these reports are the names of common models and theories：for example, one manuscript submitted to us for review flagged “the health belief model (HBM)” and “the technology acceptance model (TAM)” as instances of similarity with other literature. Many other instances involve technical descriptions; there are few alternatives to, for example, “We take the x-axis” or “Tissue sections were dehydrated through a graded ethanol series.”

Still other incidences involve common phrasing, such as “this study contributes to the … literature by offering,” which was flagged as highly similar to language in other published works, or even unavoidable phrases like “and the” or “due to.” In one such report that we reviewed, the overall score of 23% was the cumulative total of nearly 400 discrete units of “similarity,” none of which exceeded 1% and most of which were far lower than 1% (i.e., negligible).

How can AsiaEdit help?

We review the report along with the document in question, and provide a summary report to submit to reviewers/publishers refuting plagiarism where appropriate.

We offer a guarantee/certificate that your document has been checked.

If you’d like to discuss any of the issues in this article, or have a document you’d like us to check, please get in touch through [email protected].