Vocabulary entries with a self-written example sentence survive longer review intervals than bare word pairs

Hypothesis

A vocabulary log kept as a table (word, translation, entry date, review dates, grades) and reviewed on a spaced schedule such as SM-2, where a grade below 3 restarts an item's intervals and the easiness factor is floored at 1.3, will show a difference between two kinds of entries: those where the learner wrote, at entry time, one sentence of their own using the word, and those that store only the word pair. The hypothesis is that entries with a self-written sentence reach intervals of 30 days and more with fewer lapses (grades below 3 that reset the interval) and fewer reformulations than bare pairs, even though the sentence is not shown during review.

Prediction

Over several hundred entries in one log, the lapse rate at intervals above 30 days is lower for sentence entries than for bare pairs, and the difference is larger for words with several meanings or with unusual grammatical behaviour than for concrete nouns. The time spent per entry at entry time is higher for sentence entries; the prediction is that this cost is recovered in fewer repetitions before an entry reaches its first 30-day interval. If the lapse rates are equal, writing the sentence adds cost without benefit for retention as the log measures it.

Proposed test

1. Keep one vocabulary log with alternate assignment: every second new word gets a self-written sentence, the others do not; record which.
2. Review all entries with the same schedule and the same grading rule; record grade, date and elapsed interval per review.
3. After six months, compare per group: lapses per review at intervals above 30 days, mean number of reviews to reach the first 30-day interval, and entry time.
4. Inspect the words in each group for imbalance (part of speech, frequency, length) and report it.

Status

No result is claimed. A single learner's log confounds the two groups with their own motivation and with day-to-day variation; alternate assignment reduces but does not remove selection by word type. Grades are self-assigned. The hypothesis says nothing about speaking or listening skill, only about the lapse pattern in a recall log.