When universities receive a Duolingo English Test (DET) score, they see a number between 10 and 160. But behind that single number lies a carefully designed scoring system that blends cutting-edge technology with deep expertise in language assessment, psychometrics, and AI.
So, how does the DET go from your answers to your score? Here’s what you need to know.
How does the DET measure language skills?
It starts with grading—evaluating each response to determine whether it’s correct, how complete or coherent it is, or how clearly it reflects a particular skill. This happens across all item types, whether it's identifying a vocabulary word, dictating a sentence, or responding to a writing prompt.
But grading is just the first step.
Those response-level evaluations are then used to calculate your scores—quantitative measures of your English proficiency. The Duolingo English Test measures four core language skills:
- Reading
- Writing
- Listening
- Speaking
Each of these skills is scored individually to create four subscores, and together, they make up your overall score. The test also reports integrated subscores that reflect how these skills interact in real-life academic tasks:
- Literacy = Reading + Writing
- Comprehension = Reading + Listening
- Conversation = Speaking + Listening
- Production = Speaking + Writing
All scores are reported on a 10–160 scale, in 5-point increments.
How are reading and listening scored on the DET?
Most reading and listening tasks have a clear correct answer—or a small set of acceptable answers—so they don’t require subjective evaluation.
Depending on the item type, grading may involve:
- Correct/Incorrect grading (e.g., Yes/No Vocabulary)
- Partial credit grading(e.g., Interactive Listening comprehension)
- Degrees of correctness (e.g., Dictation or Highlight tasks that allow for minor variation)
Grades from individual reading and listening items are combined to form the reading and listening subscores for test takers. These item types are scored instantly and consistently based on structured rules—not machine learning models.
How are speaking and writing scored on the DET?
While most reading and listening tasks have a clear correct answer, speaking and writing require more nuanced scoring. At the end of the test, you’ll complete open-ended speaking and writing tasks—like describing a photo, responding to a prompt, or summarizing a conversation. These responses are scored using AI models trained on thousands of expert-rated samples.
These models evaluate:
- Content: Is your response relevant and well developed?
- Coherence: Is it organized and easy to follow?
- Vocabulary and grammar: Do you use a wide range of appropriate language?
- Fluency and pronunciation (for speaking): Do you speak clearly, naturally, and understandably?
The system uses automatic transcription, natural language processing (NLP), and speech processing tools to extract features from your response. A machine learning model, trained to replicate human scoring, then produces a score that aligns closely with expert judgment.
These models aren’t generic “black boxes.” They’re built around specific language constructs like intelligibility, syntactic range, and discourse cohesion, and are continuously evaluated for fairness and accuracy.
Every test adapts to the test taker
Unlike traditional language tests, the DET is computer adaptive. That means the test changes based on your performance—becoming easier or harder as needed. Put simply, if you're doing well, the test gives you more difficult items; if you're struggling, it adjusts to offer more accessible ones. This adaptivity helps the test measure your ability efficiently and accurately, in less time and with fewer questions than a fixed-form test.
The test also applies Item Response Theory (IRT) in computing test scores, which evaluates not only whether each question is answered correctly, but also how much information that question reveals about your skill level. In other words, your score is determined by both your response accuracy and the difficulty of the questions you faced. This means that even when test takers receive different sets of items, their scores remain comparable, since the test scores reflect underlying English proficiency after accounting for item difficulty.
We monitor for fairness at every step
To maintain the integrity of the test, we have several safeguards in place. off-topic or gibberish responses are flagged by AI and reviewed by proctors, and non-English responses are automatically detected and disqualified. All spoken responses are reviewed by proctors for good-faith effort.
We also conduct fairness checks across demographics—such as first language, gender, and country of origin—to ensure no group is unfairly advantaged or penalized.
[Callout box: Learn more about test fairness]
DET scores are consistent, fair, and meaningful
Scoring a high-stakes test means more than just getting the math right. It means ensuring that scores are valid; in other words, they reflect real-world academic English ability. They are also reliable, such that a test taker would receive a similar score if they retook the test under similar conditions.
Finally, DET scores are comparable: scores mean the same thing across different test takers, even if they received different questions. Behind the scenes, our Analytics for Quality Assurance in Assessment (AQuAA) system monitors all scoring data in real time to uphold these standards.
Bottom line: Smart design, not guesswork
The DET scoring system is designed to reflect how language works in real life—and to do so with speed, precision, and fairness. Every score is backed by a vast, calibrated item bank, adaptive test delivery, expert-trained AI models, and real-time monitoring and quality control.
Whether you're a student applying to university or an admissions officer reviewing applications, you can trust that each DET score is more than just a number—it’s a meaningful, valid reflection of English language proficiency.