Compilatio Magister and Magister+ enable textual analysis of submitted documents.
Depending on the version used, certain features differ:
👉 Anti-plagiarism software for teachers & universities | Compilatio
In this article, we focus on two types of detection:
- detection of exact matches and light paraphrasing, often associated with syntactic similarities
- detection of deep paraphrasing, also referred to as semantic similarities
The aim is to clarify how these two approaches work and make them easier to interpret in the analysis report.
1. Syntactic and Semantic Similarities: What They Are and How They Differ
Syntactic Similarities
(What Compilatio has always detected, now included in Compilatio Magister and Compilatio Magister+.)
Syntactic similarities concern copy-paste and light paraphrasing. These are the most common situations.
When Compilatio Magister or Magister+ detects identical or very similar word sequences between two documents, it reports a similarity. Detection thus relies directly on the words and their structure in the text.
This type of detection:
- is always active,
- requires no configuration,
- works across all analyzed languages.
It allows for the rapid identification of direct reuse, even when a few words have been modified.
Common examples:
- a passage copied word for word from a source,
- a text very close to the original, with minor synonyms or adjustments.
This detection forms the basis of the analysis. It remains essential, even when semantic similarity detection is enabled.
Semantic Similarities
(Deep paraphrasing)
Here, wording is no longer the primary factor, meaning is. A passage may be considered semantically similar even if:
- the words used are different,
- the sentence structure has been modified,
- no identical word sequence is visible at first glance.
This detection does not replace syntactic analysis. It complements it by highlighting situations that word-for-word comparison cannot always identify.
2. How Semantic Similarity Detection Works
A common question is: how can texts be compared when they do not use the same words?
The principle is simple in its overall idea. Each passage is analyzed in a way that represents its content, independently of its wording.
The comparison then focuses on the similarity of meaning between passages. When two texts express the same idea, the detection places them closer together. Conversely, different content appears more distant.
This approach relies on automated comparison. It does not correspond to human reading or pedagogical interpretation.
3. How to Configure Semantic Similarity Detection
Semantic similarity detection is configured at the administrator account level. The administrator can enable, allow, or disable this feature for users of the account.
L’administrateur peut activer, autoriser ou désactiver cette fonctionnalité pour les utilisateurs du compte.
- When semantic similarity detection is enabled, it is automatically taken into account.
- When it is allowed, users can choose whether or not to use it during analyses.
- When disabled, it is not available, and the interface specifies the reason.
This setting helps control the use of detection and avoids any ambiguity when launching an analysis.
On the user side, you can check the configured detection types and adjust them if necessary.
4. How to Read and Interpret Results in the Analysis Report
The analysis report presents several distinct elements. Its purpose is to highlight similarities, rather than to provide an automatic conclusion.
The Overall Score
It reflects all the similarities detected in the document and takes into account:
- syntactic similarities,
- semantic similarities, when this feature is enabled.
These two types of similarities appear separately to facilitate reading.
Sources and Points of Interest
In the list of sources, some may be associated with semantic similarities. Visual indicators make them easy to identify.
Points of interest highlight the relevant passages in the document. They serve as reading guides, without constituting a judgment in themselves.
The Side-by-Side View
The side-by-side view allows you to compare the analyzed document with a source.
- bold passages correspond to exact similarities,
- non-bold passages indicate semantic reformulations.
Both may appear within the same excerpt. This is normal and reflects how the content has been reformulated.
5. Paraphrasing and Translation: Two Distinct Cases
Semantic similarity detection may resemble translation detection. However, these are two different situations.
Reformulation refers to a text rewritten in the same language, using different wording but conveying equivalent content. Translation, on the other hand, involves rendering a text from one language to another while preserving the same meaning.
In Compilatio Magister+ reports, these two types of similarities are displayed separately to facilitate reading and avoid confusion.
6. Why Some Similarities Are Not Measured
To ensure reliable results, semantic similarity detection relies on defined thresholds. Compilatio highlights similarities that are pedagogically or analytically relevant, not all possible similarities.
This approach ensures the report remains readable and helps avoid over-interpretation. As such, some similarities may not be detected. This is not an error but the normal behavior of the system.
Semantic detection is applied after the initial analysis and only to primary sources. A minimum number of shared words is required to trigger the analysis; below this threshold, the source is not considered.
In practice, a heavily reworded passage with no shared vocabulary, or one found only in secondary sources, may not appear in the results.
This methodological choice ensures a balance between precision, readability, and relevance.
7. Key Takeaways
- Syntactic similarities are based on words,
- semantic similarities relate to meaning,
- semantic detection complements the existing analysis, it does not replace it,
- results highlight areas to review, not conclusions,
- the report must always be interpreted in its context.
This article has been automatically translated. If you notice a translation error, please contact us.