Retroflex collapse and tone neutralisation in an HSK 2 evening class

The two patterns have different causes. The zh/ch/sh → z/c/s collapse is consistent with L1 transfer: English has no retroflex series, and the learners' dental [ts]/[s] are close enough to be perceptually acceptable to themselves. The tone neutralisation in connected speech, despite correct isolated production, points to a processing problem rather than a motor one: learners allocate attention to segmentals and lexis, and tone 2/tone 3 contours are harder to discriminate and stabilise when coarticulated.

Priority should go to tones in connected speech. First, functional load: tone errors produce many more lexical confusions in real communication than a dentalised zh, which context usually repairs. Second, prevalence: 11/18 show the tone problem in free talk, and it affects the four-week speaking test directly. Third, teachability: the learners already produce the contours in isolation, so the gap is transfer to connected speech, which short, distributed practice can address. Retroflex work should continue in parallel as perceptual discrimination and minimal-pair production, but it should not consume the whole 15 minutes.

A four-phase, 15-minute routine: (1) 3 minutes of perceptual discrimination - teacher says pairs such as 十四/四十, 买/卖, 汤/糖; students raise cards or point, with no production pressure. (2) 4 minutes of controlled production - tone-pair chanting in short frames (hěn máng, hěn mǐn, hěn màn) and retroflex/dental minimal pairs in carrier phrases, with immediate but brief feedback. (3) 5 minutes of contextualised use - a 4-line dialogue or information-gap task that forces the target tones and the zh/ch/sh series in meaningful exchanges, e.g., asking prices and quantities. (4) 3 minutes of recycled recall - two sentences from the previous lesson, retrieved without written pinyin.

The sequence moves from perception to controlled production to contextual use, reflecting the claim that learners must first notice and discriminate a contrast before they can stabilise it in production. It is distributed rather than massed: 15 minutes twice a week for four weeks gives eight short cycles, which is more effective for automatisation than one long clinic. It also embeds the targets in the lesson's communicative content, so learners do not treat pronunciation as a separate subject.

The teacher should not remove pinyin abruptly, because that would raise anxiety and reduce the comprehensible input the beginners need. Instead, pinyin should be gradually withdrawn as a reading support. Practical steps: first, present new vocabulary with characters and pinyin, but do all review and recycling with characters only; second, use a two-column routine in which students match spoken words to characters before seeing pinyin; third, set short character-recognition quizzes on the 30–40 highest-frequency characters in the unit, so the goal is recognition rather than handwriting. For the speaking test, pinyin can remain on the prompt card but not on the assessment criteria, so the test measures speech, not reading. The underlying reasoning is that pinyin dependency is an orthographic-processing issue, not a moral one: learners need many successful encounters with characters in meaning-focused tasks before the visual form becomes automatic. Four weeks is enough to build recognition of a limited set, but not to eliminate pinyin entirely, so the teacher should set a realistic target and tell students what it is.