A student providing a multi-word response in a computerized language teaching system provides a manual input concurrently with each responsive word. For example, he might enter a keystroke correspondent to the first letter of each word. When using the teaching computer silently, a student will typically "speak" each word mentally as he enters a keystroke, so the limited experience is almost as effective as speaking out loud. When a student types one or more keystrokes concurrently with each word that he speaks, the computer will be able to detect when a student is responding with a correct word, but merely mispronouncing it. Also, since the computer will receive a keystroke as the student starts each new word, it is better able to distinguish the boundaries between words and recognize them more reliably.