When your child finishes a practice session on PSLEPrep, they see a scorecard with specific marks for each dimension — pronunciation, fluency, content, vocabulary. But where do those numbers come from? Are they reliable? And how do they relate to the actual PSLE exam?
This article explains how PSLEPrep scores your child's oral practice, why we chose this approach, and what the scores mean for PSLE preparation.
Two Types of Scoring for Two Types of Skill
When your child speaks, there are two distinct things to assess:
- How it sounds — Are the tones correct? Is the pronunciation clear? Is the delivery smooth?
- What was said — Did they answer the question with depth? Did they give reasons and examples? Did they use appropriate vocabulary?
These require fundamentally different assessment methods. You cannot judge pronunciation from a written transcript — you need to hear the actual audio. Conversely, you cannot judge content depth from sound waves alone — you need to understand the meaning of the words.
PSLEPrep uses a specialised audio analysis engine for how it sounds, and an AI language model for what was said. Each does what it is best at.
Pronunciation and Fluency: Scored from Audio, Not Guessing
When your child reads a passage aloud, PSLEPrep sends the recording to a Chinese speech evaluation engine that analyses pronunciation at the character level. For every single character in the passage, it checks:
- Was the correct tone used? (Mandarin has 4 tones plus neutral)
- Was the correct pinyin (consonant + vowel) pronounced?
- Was the character skipped, or did the student add extra words?
This is not a transcript comparison — it is acoustic analysisof the actual sound waves. The engine compares your child's audio against the expected pronunciation of each character, including the specific tone. This is the same type of technology used by language learning platforms worldwide for Mandarin pronunciation assessment.
The engine is calibrated for upper-primary students — it expects the level of a P5–P6 child, not a native adult speaker. It returns detailed scores for accuracy, fluency, and standard delivery. PSLEPrep then maps these to the PSLE scoring dimensions using fixed, consistent rules.
Why this matters
Many AI tools score pronunciation by reading a transcript — essentially guessing from text. But a transcript cannot tell you whether your child said 是 with the correct 4th tone or flattened it to the 1st tone. PSLEPrep analyses the actual recording, so tone errors — the most common weakness for Singapore students — are caught and scored accurately.
Content and Vocabulary: Scored by AI Language Understanding
For the conversation component, content depth and vocabulary are assessed by an AI language model — the same type of technology behind ChatGPT. After your child completes the three-question conversation, the AI reviews the full transcript and assesses:
- Content: Did your child give reasons? Provide examples? Elaborate beyond one-line answers?
- Vocabulary: Did they use topic-appropriate words? Did they use connectors like 因为…所以… and 虽然…但是…?
The AI also writes personalised bilingual feedback — in Chinese and English — explaining what your child did well and what to work on next. This includes specific suggestions like which connector to try next time, and a tip you can practise at home.
How PSLEPrep Scores Map to the Actual PSLE
PSLEPrep uses the same scoring dimensions and mark allocations as the PSLE Chinese Oral exam. Here is how they line up:
Reading Aloud (朗读) — 20 marks
| Dimension | Marks | How PSLEPrep Scores It |
|---|---|---|
| 语音 Pronunciation | 7 | Audio analysis — tone accuracy (50%) + pinyin accuracy (50%) |
| 准确 Accuracy | 4 | Audio analysis — skipped, added, or substituted characters detected |
| 流利 Fluency | 5 | Audio analysis — smoothness, pacing, hesitation |
| 语感 Expression | 4 | Audio analysis + AI adjustment for content-appropriate emotion |
Conversation (会话) — 30 marks
| Dimension | Marks | How PSLEPrep Scores It |
|---|---|---|
| 内容 Content | 10 | AI language analysis — depth, reasoning, examples, structure |
| 词汇 Vocabulary | 8 | AI language analysis — range, connectors, topic-appropriate terms |
| 发音 Pronunciation | 6 | Audio analysis — per-turn tone and pinyin accuracy |
| 流利 Fluency | 6 | Audio analysis — smoothness of delivery across all answers |
Notice that in conversation mode, content and vocabulary together account for 60% of the marks (18 out of 30). This reflects what PSLE examiners actually prioritise — it is the depth of your child's answers, not just how they sound, that determines whether they reach AL1.
Consistent Scoring You Can Track Over Time
PSLEPrep uses fixed scoring rulesfor pronunciation and fluency. The same audio performance will always produce the same score — there is no variation from session to session. This means when your child's scores improve, the improvement is real and measurable.
Content and vocabulary scores involve AI judgement (because evaluating answer depth requires understanding meaning), so there can be slight variation. But the scoring guidelines are strict: the AI follows specific criteria for each dimension, with clear caps to prevent over-scoring.
Every session's scores are saved to your child's history. You can track progress over time and see exactly which dimensions are improving and which need more work.
What Happens When Audio Analysis Is Unavailable?
Occasionally, the audio analysis service may be temporarily unavailable. When this happens, PSLEPrep does not guess at pronunciation, fluency, or expression scores. Instead, those dimensions are shown as “Not scored” with an explanation. Only the accuracy dimension (which can be checked by comparing the transcript against the passage) is scored.
We believe this is more honest than showing inflated or unreliable numbers. If you see “Not scored” on any dimension, simply try again — full scoring will resume once the audio analysis service is back online.
What Your Child's Scores Mean in Practice
PSLEPrep displays a Chinese band label alongside every score to give you quick context:
| Band | Score Range | What It Means |
|---|---|---|
| 上 | 90%+ | Excellent — AL1 performance level |
| 中上 | 75–89% | Good — solid AL2 performance |
| 中 | 60–74% | Average — typical AL3 range, room for improvement |
| 中下 | 45–59% | Below average — focus on the weakest dimension first |
| 下 | Below 45% | Needs work — but regular practice will show improvement quickly |
The most useful thing about PSLEPrep's scoring is not the overall number — it is the dimension breakdown. If your child scores 15/20 on reading but only 2/5 on fluency, you know exactly what to practise. If they score 20/30 on conversation but only 4/10 on content, the answer is clear: work on giving longer, more detailed answers using the P.E.E. framework.
PSLEPrep gives your child specific, actionable feedback after every practice session — scored the same way PSLE examiners assess. Start free trial →
Frequently Asked Questions
How does PSLEPrep score pronunciation?
PSLEPrep analyses the actual audio recording using a specialised Chinese speech evaluation engine. It checks every character's tone and pinyin against the expected pronunciation — this is acoustic analysis of the sound waves, not a transcript comparison. The engine is calibrated for upper-primary students (P5–P6 level).
Are PSLEPrep scores the same as PSLE scores?
PSLEPrep uses the same scoring dimensions and approximate mark allocations as the PSLE Chinese Oral exam (reading /20, conversation /30). However, we are not SEAB examiners. Our scores are designed to be a reliable indicator of PSLE-level performance and to track improvement over time. SEAB does not publish exact rubric weightings, so our allocations are based on published syllabi and expert analysis.
Why do pronunciation scores sometimes seem strict?
Our audio analysis engine scores tones strictly because tone accuracy is the most impactful pronunciation skill for PSLE. A student who sounds 'okay' to a non-Chinese-speaking parent may have systematic tone errors that a PSLE examiner would catch. The strict scoring is intentional — it highlights the specific errors your child needs to fix, and ensures improvement is visible when they practise.
What does 'Not scored' mean on the scorecard?
If you see 'Not scored' on pronunciation, fluency, or expression, it means the audio analysis service was temporarily unavailable for that session. Rather than guessing, PSLEPrep only shows scores it can calculate reliably. Try again and full scoring will resume.
How is the conversation content score calculated?
Content is scored by an AI language model that reviews the full conversation transcript. It assesses whether your child answered each question with depth (reasons, examples, personal connections), covered the key elements for each question type (describe, opinion, experience), and spoke at sufficient length. Content is worth 10 out of 30 conversation marks — the single largest dimension — reflecting PSLE examiners' emphasis on answer depth over pronunciation alone.