A TOEIC score of 780 can sit beside complete silence in a Monday meeting. The same learner may talk for twenty minutes with a close colleague on Friday night. Speaking confidence lives partly in the situation: the partner, topic, stakes, preparation time and expected response all change the decision to speak. A course that measures confidence with one general label misses the most useful question. We need to know where the learner speaks, and where the voice disappears.
Readiness is momentary
Researchers call this willingness to communicate, often shortened to WTC. The term describes a learner's readiness to enter a conversation at a particular time with particular people, wording that turns “quiet” from a permanent personality judgement into something a teacher can investigate.
Lee and Chiu modelled responses from 1,269 Korean secondary and university learners. Lower anxiety was associated with greater willingness across classroom, out-of-class and digital settings. The paths differed by setting, though. Self-perceived ability was tied to face-to-face anxiety, while an ideal future English-speaking self had a different relation to anxiety in digital communication. This was a survey model, so it cannot show that lowering anxiety by itself will make someone speak. It does show why one confidence score is too blunt.
Eight words are enough here: who, where, why, with whom, and at what cost.
The partner changes the task
Taiwanese learners often receive advice to “speak more English,” an instruction that hides several tasks inside one phrase because asking a classmate where they ate lunch carries little social cost while correcting a senior manager's figure during a video call carries much more. Chatting with a voice assistant removes facial reaction and status. Each event asks for different language and a different tolerance for uncertainty.
Tzu-Yu Tai's 2024 study in Computers & Education followed 92 first-year university students in Taiwan for twelve weeks. Thirty-one interacted with Alexa or Google Assistant, twenty-nine with first-language English speakers, and thirty-two with other second-language speakers. They completed two roughly ten-minute interactions each week. The voice-assistant group reported larger gains in WTC than the two human-partner groups. Interviews linked the result to convenience, familiarity and the absence of human judgement, among other factors.
The result supports a useful staging choice. A low-stakes machine partner can provide early speaking turns. It cannot supply the timing, repair, facial cues or relationship work of a human conversation. A voice assistant is a rehearsal room. The meeting remains a different room.
Build a situation map
Before assigning extra conversation practice, write down four recent speaking situations. Keep the descriptions narrow: answering a teacher in a four-person online class, ordering at a café near Zhongxiao Fuxing, giving a 90-second project update, or joining an English board-game event with two strangers. For each one, record three things after the attempt: whether the learner entered the conversation, how long the first turn took to begin, and what stopped the next turn.
The pattern is more useful than a rating such as “confidence: 6/10.” A learner may begin quickly with strangers yet avoid disagreement. Another may answer every direct question and never volunteer a topic. Those are different lesson-design problems.
This also changes feedback. “Be more confident” gives no next action. “Use one entry phrase before the agenda moves on” does. Start there. For a project update, that phrase might be “Can I add one detail about the deadline?” At a social event, the learner could mention how they know the host. The phrase buys entry. The learner still has to listen and respond.
Change one variable at a time
One sensible sequence holds most of the situation steady while changing one demand. The learner first explains the same work problem to a voice recorder, then to a familiar classmate, then to a classmate who must ask two follow-up questions. A later version changes the audience to a teacher. The topic stays familiar while the social load rises.
Partner rotation needs the same care. Sudden weekly reshuffling can keep anxious learners busy managing introductions. Three encounters with one partner can establish a usable rhythm. The fourth encounter with someone new tests whether that rhythm travels. For paired tasks, shared control matters: both people need information, decisions and room to initiate.
Here is my preference: keep the speaking goal small enough to observe. “Contribute twice before minute twelve” beats “participate actively.” One target names two turns and a deadline. The other invites a vague judgement after class.
Measure transfer, not comfort alone
Comfort can help a learner accumulate turns, yet comfort can also become a protected corner. If every successful exchange happens with the same friend, the course has evidence of one safe relationship. It has little evidence of transfer.
Every few lessons, reuse the language under a modestly changed condition. Move from a prepared answer to one follow-up question. Move from a peer to a teacher. Move from text chat to audio, then from audio to camera. Record the point where participation drops. That point sets the next practice task.
For one month, a transfer ladder can stay narrow: explain one delayed project to a recorder, a familiar peer, an unfamiliar peer and a teacher, then give the same explanation during a timed group update. Keep the message and entry phrase stable while the audience and response pressure change. Five successful recordings alone show control in recording. One workable group turn shows that part of the control survived a new condition.
Some learners face consequences that classroom design cannot remove. A junior employee may have good reason to stay cautious around a director. A teenager may avoid public error after classmates have mocked an accent. The teacher should never treat silence as a defect detached from power and history. The teachable part is preparation, entry language, repair language and a graduated set of safer attempts.
Tomorrow, choose one real situation and alter one variable. Keep the topic. Change the partner. Count the seconds before the first turn.