Reaction Time as a Fluency Metric: What Speed of Recall Actually Predicts

Ask anyone who has actually become fluent in a second language as an adult what changed, and almost none of them will say "I learned the words." They knew the words for years. What changed was that the words started arriving on time — fast enough to keep up with a conversation, fast enough to interrupt themselves, fast enough not to get smiled at and switched to English at the café counter. Fluency, the thing the word is actually pointing at, is overwhelmingly a question of speed.

This is the core observation Word Exchange Plaza is built around. The plaza does not just track whether you got an item right. It tracks how long you took, and that timing data feeds back into the spaced-repetition schedule the same way accuracy does. Words you answered fast and correctly fade. Words you answered correctly but slowly come back, because slow correct is the leading indicator of about-to-be-wrong. This essay is about why that distinction matters, what the cognitive science actually says about it, and how to train for fast recall whether you use the plaza or not.

Why "Right Eventually" Is the Wrong Bar

Almost every popular language app uses correctness as the primary feedback signal. You see a word, you select an answer or type a response, the app tells you whether it matches. Time pressure, when present at all, is usually a soft cue (a streak bonus for fast answers, a slight color change on a slow one) rather than a first-class metric. The implicit message is that the goal is to be right, and that being right slowly is functionally the same as being right quickly.

In a real conversation, this is false in a way that becomes obvious within thirty seconds. The other person asks you a question. You start the mental process of finding the right Spanish verb. They wait. You find it after four seconds. You produce the correct answer. They have already inferred from your hesitation that the conversation will be exhausting for both of you, and the next sentence is in English. You were right. You were too slow. Right and too slow is functionally the same as wrong, in the only context that matters.

The cognitive science behind this is well-established. Reaction time on lexical retrieval tasks is one of the most-studied measures in psycholinguistics, and the consistent finding across decades of research is that response latency on a known word tracks the strength of its memory representation more sensitively than accuracy does^[1]. By the time accuracy starts to fail, the word has already been weakening for a long time. Reaction time catches the decline early.

The Recall-Speed Curve

What does that decline actually look like? In controlled settings, the curve has a fairly consistent shape. A freshly-learned word starts at a high latency — it takes you a couple of seconds to retrieve it the first few times you encounter it. With repetition, the latency drops sharply, then asymptotically approaches a floor that is roughly the same as your reaction time on words in your native language for that meaning category. A native English speaker's reaction time to recognize "apple" sits in the range of 500–700 milliseconds. A learner's reaction time to recognize the equivalent in Spanish or Hindi might start at 3,000+ ms and, with enough exposure, settle into 700–1,200 ms.

The interesting region is the transition between "I can produce the word if you give me time" and "the word arrives fast enough not to interrupt the conversation." Most adult learners stall in the first half of that transition. They acquire vocabulary at a reasonable rate, push the recall latency from 5 seconds down to about 2 seconds, and stop. Two seconds is enough to pass a written quiz. It is too slow to keep up with even a slow native speaker. That gap is where the recognition-but-can't-speak plateau lives.

Closing the gap requires deliberately training the speed component, which means putting yourself in conditions where slow correct answers are flagged as a problem and re-tested. You can do this with a stopwatch and a vocabulary list — the technique is older than the smartphone — but it is much easier to do with a piece of software that records the timing automatically and adjusts what you see next based on it.

What the Memory Research Actually Shows

The connection between recall speed and memory strength was formalized by John Anderson's ACT-R model of memory in the 1980s and has been refined steadily since. The short version: every time you successfully retrieve a piece of information from memory, you make it slightly easier to retrieve next time, both faster and more reliably. The strength of a memory item, in ACT-R terms, can be inferred from the latency of successful retrievals — high-strength items are retrieved fast, low-strength items are retrieved slowly even when retrieval succeeds^[2].

This has direct implications for how spaced-repetition systems should work. Classic spaced-repetition algorithms (Anki's SuperMemo-2 algorithm, for instance) use a self-rated difficulty score after each review — you tell the app "easy / good / hard / again," and the next review interval is computed from that. Self-rating is noisy. People are bad at telling the difference between "I knew this comfortably" and "I knew this but it took me four uncomfortable seconds." Latency is a much better signal because it is measured automatically and doesn't depend on the learner's introspective accuracy.

This is part of why newer SRS systems — including the one inside Word Exchange Plaza — increasingly use response latency as a first-class scheduling input rather than just an optional metric. A correct answer that took 4 seconds is treated as a borderline result and the item comes back sooner than a correct answer that took 1 second on the same item. Over enough repetitions, this drives the learner's average latency on the deck down toward the native-speed floor instead of letting it stabilize at the "right but slow" plateau.

What "Fluent Speed" Actually Looks Like

It is worth being concrete about what reaction times we are talking about. In a normal conversation between native speakers of any language, the average gap between one speaker finishing and the next speaker starting is around 200 milliseconds — a remarkably consistent finding across more than a dozen languages, including ones as structurally different as English, Japanese, Yélî Dnye, and ‡Ākhoe Hai‖om^[3]. That 200 ms is the gap. The actual mental work — parsing what the other person said, formulating a reply, beginning to speak — is happening in parallel with the end of their sentence, because there is not enough time to do it serially.

This implies something important. To converse at native pace, your individual word recall latencies have to be well below the 200 ms gap, because the recall has to happen alongside other processing (sentence parsing, social inference, etc.) and not occupy the whole budget. A learner whose word recall latency is even 1,500 ms — fine for a quiz, fine for an app's "correct" indicator — has no spare cognitive capacity to do anything else. Conversation breaks down not because they don't know the word, but because knowing the word at 1,500 ms means there is no room for the rest of the cognitive work a real exchange requires.

The practical implication for training is that the goal is not "answer correctly." The goal is "answer correctly within roughly twice your native-language baseline." For most adults, that is roughly the 600–1,200 ms band. Anything slower than that is a word you have not yet built fluency on, regardless of whether you can produce it given enough time.

How to Train for Speed

Whether or not you use Word Exchange Plaza, four practical techniques will improve your recall speed on existing vocabulary without requiring you to learn new words.

1. Use a timer when you drill. If your current SRS does not enforce a time limit on responses, set one yourself. Three seconds for a single-word response is a generous starting limit; tighten to two seconds once that becomes comfortable, then to one. The friction of the timer is exactly the point — it surfaces the words that have decayed into "right but slow" and pushes them back into your active recall pool.

2. Drill out loud. Verbal production is slower and noisier than mental retrieval, so a verbal drill exposes weaknesses that a mental drill hides. The plaza enforces this by being voice-driven by default. If you are using Anki or a similar tool, train yourself to say the answer out loud before flipping the card, and count any answer where you hesitated as a "wrong" for scheduling purposes.

3. Shadow native audio. Speech shadowing — repeating a native speaker's audio in real time, with as little delay as possible — is one of the most effective ways to train recall speed because it forces you to operate at native cadence. Even ten minutes a day of shadowing a podcast you mostly understand will measurably tighten your latencies within a few weeks.

4. Use the words in self-talk. Narrating your day in your target language, out loud, forces vocabulary to come at conversational speed because the next thing you want to say is right behind the current word. The first week of this is humbling. The fourth week is transformative.

An Honest Caveat: When Speed Is Not the Goal

Reaction time is the right thing to optimize for if your goal is conversational fluency. It is not the right thing to optimize for if your goal is, say, reading literary fiction in your target language, or writing a research paper, or doing high-precision technical translation. Those tasks reward depth of vocabulary and grammatical accuracy more than they reward speed. A learner whose vocabulary is deep and slow may be exactly the right learner for a literary translation job and exactly the wrong learner for a phone call with a contractor.

Most adult learners, in our experience, want both — but more often than not, they have been training only for the depth half (because that is what most apps measure) and have completely neglected the speed half. The plaza's bet is that adding the speed half is the highest-leverage thing most adult learners can do. Reaction time as a metric is the most direct way to do it.

The Short Version

Recall accuracy tells you whether the word is in your head. Recall speed tells you whether it will arrive in time to be useful. Real conversations care about the second one almost exclusively. The cognitive science has known this for forty years. The training tools are finally catching up.

If you are stuck at the "I know all the words but I can't speak" plateau, adding a speed-aware drill to your routine — whether that is the plaza, a stopwatch and Anki, or shadowing podcasts on the walk to work — is the single most reliable way to climb out of it. The vocabulary will not change. The speed at which you can reach for it will. That is the change that turns a learner into a speaker.