Which English Sounds Are Hardest for Chinese TOEFL Learners?

Phương pháp nghiên cứu

Otherwise the method matches the main study: each recording is scored by automatic pronunciation assessment (0–100 per phoneme), and word-final -s is recovered by aligning the target to the recognised words. A sound is ranked within a group only when it accumulates roughly 1,000+ measurements in that group; rarer sounds are left unranked. We report denominators, means, and group differences, but no significance tests — repeated learners and prompts are not independent observations.

Mẫu

6,057 of 23,390 TOEFL 2026 Listen and Repeat sessions set a LingoLeap interface language: 4,904 Chinese-interface sessions (32,231 scored attempts) and 1,045 English-interface sessions (6,879 attempts). The number of distinct learners is not recorded.

Thời gian

From 1 January 2026. The end of the extraction window is not recorded in this study's data file.

Nguồn

A read-only, offline aggregation of LingoLeap practice records, joined to each learner's interface-language setting, which stands in for first language. Aggregates only: no recordings, transcripts, user IDs or individual attempts.

Cách tạo điểm số

Phoneme accuracy (0–100) comes from automatic speech pronunciation assessment; sentence scores (0–5) are LingoLeap AI platform scores. None of these are ETS scores.

Loại trừ

The 17,333 sessions with no interface language are outside the comparison. A sound is ranked within a group only with roughly 1,000+ measurements in that group, so /θ/ is unranked for English-interface learners.

Hạn chế

Interface language is a proxy, not a declared first language, and covers only about 26% of sessions. Observational: no significance tests, and repeated learners and prompts are not independent. Pronunciation is judged by a machine, not a phonetician.

Pattern 1: The one sound that actually diverges is /θ/

/θ/ is the hardest sound for the Chinese-interface group at 63.5/100; for the English-interface group it does not rank as a difficulty at all.

<figure style="margin:0;max-width:660px">
  <img src="https://lingoleap.ai/research/charts/chinese-learners-phoneme-compare.svg" alt="Figure 1. Accuracy by sound, both groups" width="660" height="472" style="max-width:100%;height:auto" loading="lazy">
  <figcaption>Nguồn: <a href="https://lingoleap.ai/research/chinese-learners-toefl-pronunciation">LingoLeap, Which English Sounds Are Hardest for Chinese TOEFL Learners?</a></figcaption>
</figure>

The size of the gap is worth pausing on. The Chinese group’s second-hardest sound, /ŋ/ (72.5), already sits about 9 points above /θ/. So /θ/ is not “one more slightly hard sound” — it is a single outlier, cleanly separated from the rest. Of its 3,252 measured occurrences, 36.0% fell below the 60-point threshold.

Pattern 2: Almost everything else is the same

This section is really the control for the last one. If the two groups differ only at /θ/ and match everywhere else, the /θ/ gap is more likely a real linguistic effect than an artefact of sampling or overall level.

Sentence length behaves almost identically too. Reproduction falls as sentences grow in both groups, and the curves nearly overlap — 79.5% vs 81.2% on the shortest sentences, 51.5% vs 53.2% on the longest.

<figure style="margin:0;max-width:660px">
  <img src="https://lingoleap.ai/research/charts/chinese-learners-sentence-length.svg" alt="Figure 2. Reproduction by sentence length, both groups" width="660" height="320" style="max-width:100%;height:auto" loading="lazy">
  <figcaption>Nguồn: <a href="https://lingoleap.ai/research/chinese-learners-toefl-pronunciation">LingoLeap, Which English Sounds Are Hardest for Chinese TOEFL Learners?</a></figcaption>
</figure>

Word-final -s tells the same story: 10.7% dropped in the Chinese-interface group versus 9.6% in the English-interface group. Put together, the picture is clear — the two Listen and Repeat profiles largely overlap, and the one meaningful crack is at /θ/.

How to read this

The largest group — learners who set no interface language — also ranks /θ/ hardest, at about 61, matching the Chinese-interface group. Given LingoLeap’s heavily Chinese-speaking user base, that is consistent with the unset group being largely Chinese-first-language. It corroborates the pattern but is not independent confirmation.

For a Chinese-first-language test-taker, this narrows a vague worry (“my pronunciation isn’t good enough”) into a concrete target: /θ/ is worth isolating and prioritising, while on most other sounds your gap to English-interface learners is small. It also argues for targeting the few sounds that actually diverge by first language, rather than “practising pronunciation” in general.

What this study cannot show

• Interface language is a proxy for, not the same as, first language; it cannot separate heritage speakers, bilinguals, and similar cases.
• Limited coverage: only about 26% of learners set a UI language, and the English-interface group has just 1,045 sessions — a much smaller sample than the Chinese group.
• /θ/ is unranked for the English-interface group because its measurement count fell below the threshold, so we say only that it does not enter the difficulty ranking — not that it is easy for English-interface learners.
• These are automatic-assessment outputs, not phonetician judgements; the /θ/ substitution mechanism (/s/ vs /t/) is not tested directly here.
• The design is observational: no significance tests and no causal claims.

Turn the finding into practice

If you are a Chinese-first-language learner, the most practical next step is to train /θ/ inside real sentences, not in isolation. These two entry points cover the method and the practice:

Beyond pronunciation, the interview companion study finds Chinese- and English-interface learners score almost identically on Delivery, Language Use, and Topic Development.

Dữ liệu

Cách trích dẫn

Notes and sources

1. [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.
2. [2] Saito, K. (2019). Effects of second language pronunciation teaching revisited. Language Learning, 69(3), 652–708.
3. [3] Microsoft. (2026). Use pronunciation assessment. Microsoft Learn.

Questions we hear from learners