Companion study · first-language comparison
Do Chinese Learners Score Differently on TOEFL Speaking Interviews?
LingoLeap Research Team · September 2026
Quick answer
Comparing 3,476 Chinese-interface learners with 746 English-interface learners on Take an Interview, the two groups are close on nearly every dimension: the delivery trio (fluency/prosody/pronunciation) differ by at most about 1.4 points, grammar/vocabulary (0–5) is 3.11 vs 3.13, elaboration 3.26 vs 3.28, and their score distributions nearly overlap. Under this (weak) interface-language proxy, there is no detectable first-language gap on interviews.
This is the first-language companion to the TOEFL Speaking scoring series. Many Chinese-first-language test-takers worry they are “naturally disadvantaged” at speaking. We run the one first-language cut the interview data supports and ask whether that worry holds up — with the caveat that interface language is only a proxy, so the result must be read alongside its limits.
The data, and one important caveat
Learners did not declare a first language. We use their chosen LingoLeap interface language (Chinese vs English) as a proxy, joined by user. Three consequences: interface language is not native language — many English-interface users may still be Chinese-first-language; only about 27% of learners set a usable UI language (4,222 of 15,512 sessions); and the English-interface group is small (746). This page therefore describes differences between interface groups, which is not the same as native-language differences and does not represent all test-takers.
Otherwise the method matches the rest of the series: each answer is scored by LingoLeap AI, its signals mapped to the three public dimension names (not ETS scoring), aggregated over bands 2.0–4.5. We report each group’s means and the difference, with no significance tests.
Pattern 1: The two groups are almost identical
Dimension by dimension, the Chinese- and English-interface groups differ by small margins — none opens a meaningful gap.
| Dimension | Chinese UI | English UI | Difference |
|---|---|---|---|
| Fluency (0–100) | 70.1 | 68.7 | +1.41 |
| Prosody (0–100) | 76.5 | 76.2 | +0.33 |
| Pronunciation (0–100) | 79.3 | 78.3 | +1.05 |
| Speech rate (wpm) | 113 | 111 | +2.04 |
| Answer length (words) | 77 | 75 | +1.46 |
| Grammar/vocab (0–5) | 3.11 | 3.13 | -0.02 |
| Elaboration (0–5) | 3.26 | 3.28 | -0.02 |
| Topic relevance (0–5) | 3.82 | 3.83 | -0.01 |
| Complex-sentence share | 37% | 37% | +0.0 pp |
› A closer look
The differences are not only small but inconsistent in direction (Chinese-UI is marginally higher on some, lower on others), which reads more like noise than a stable gap. Importantly, both groups are plausibly Chinese-heavy (the English-UI group may include Chinese-first-language users), so part of this closeness may simply be that the two groups have similar first-language makeup.
Pattern 2: The score distributions nearly overlap
If the groups genuinely differed in ability, their score distributions should be visibly offset. Instead the two nearly coincide, both concentrated in the 3.0–3.5 bands.
› A closer look
The overlapping distributions and the near-equal per-dimension means corroborate each other: with each dimension close, the combined score distribution follows. This cannot prove there is no difference, but in this data and under this proxy, interview scores show no first-language stratification.
How to read this
For a Chinese-first-language test-taker, this narrows a vague worry (“I’m naturally bad at speaking”): on the interview task there is no evidence that Chinese-first-language learners are systematically separated on Delivery, Language Use, or Topic Development. The clearer first-language pattern is in pronunciation — see the Listen and Repeat pronunciation study, where /θ/ is the single hardest sound for Chinese-first-language learners.
In other words, interview scores look more like a general skill everyone trains (complete, coherent, substantive answers) than something stratified by first language. Rather than worrying about native language, train the signals that differ most across bands, in the Delivery and Language Use chapters.
What this study cannot show
- Interface language is a proxy for, not the same as, first language; the English-interface group may contain many Chinese-first-language users, masking a real difference.
- Limited coverage: only about 27% of learners set a UI language, and the English-interface group has just 746.
- Near-equality is not proof of no difference: a real gap could be hidden in a small sample (no significance tests).
- Scores are LingoLeap AI scores, not official ETS results; dimension names borrow the public framework without reproducing rubric text.
- Covers the interview task only; first-language patterns may differ on other speaking tasks (e.g. Listen and Repeat).
Dataset note
LingoLeap Research Team. (2026). Do Chinese Learners Score Differently on TOEFL Speaking Interviews? LingoLeap Research, version 1.0.
https://lingoleap.ai/research/chinese-learners-toefl-speaking
Sample: Chinese-interface 3,476 learners / 13,045 answers; English-interface 746 / 2,771.
Notes and sources
- [1] Best, C. T., & Tyler, M. D. (2007). Nonnative and second-language speech perception. In Language Experience in Second Language Speech Learning (pp. 13–34). John Benjamins — first-language effects are strongest at the phoneme level, weaker at the discourse level.
- [2] LingoLeap Research. Which English sounds are hardest for Chinese TOEFL learners? — the companion pronunciation study on the Listen and Repeat task.
- [3] ETS. TOEFL iBT Speaking — the public dimension names, used only as an organizing framework.
Questions we hear from learners
Are Chinese speakers at a disadvantage on TOEFL Speaking interviews?
Not visibly in this data. Chinese-interface and English-interface learners score almost identically across Delivery, Language Use, and Topic Development, and their score distributions nearly overlap. This is an association under a weak proxy (interface language, not verified first language), so it does not prove there is no difference — but it gives no sign of a systematic first-language gap on interviews.
How was first language determined?
It was not declared. We use the learner's LingoLeap interface language (Chinese vs English) as a proxy, joined by user. This is imperfect: many English-interface users may still be Chinese-first-language, and only about a quarter of learners set a UI language, so the comparison describes those interface groups rather than verified native languages.
Where do Chinese learners actually differ on TOEFL Speaking?
On the interview task in this data, not much across the scored dimensions. A clearer first-language pattern shows up in pronunciation on the Listen and Repeat task, where /θ/ (the 'th' sound) is the single hardest sound for Chinese-first-language learners — see the companion pronunciation study.
Are these official ETS scores?
No. These are LingoLeap AI scores mapped to the public dimension names, from real practice answers — not official ETS results.