The Rochester projectSummarising Dickens, and checking what survives
Split the chapter into sentences, score them, keep three, then test the summary against five facts.
The learner takes the first chapter, "In the Old City of Rochester", and splits it into sentences: 101 of them with at least five words, averaging 33 words each, the longest 174. Two early bugs appear. The chapter heading is glued to the first sentence, and the fact checker, looking for "Watts", misses "RICHARD WATTS" in capitals until it ignores case. Then the summariser counts how often each meaningful word appears, leaving out common words such as "the" and "and". The top words are travellers, 21 times, and Watts, 12. Each sentence gets a score from its words, and the three highest-scoring sentences form the summary.
Three-sentence summaries of Chapter I, checked against five key facts, our Python run, 28 September 2026| Method | Summary length | Facts kept (of 5) | Facts kept |
|---|
| Add up word scores | 289 words | 3 | Watts, six travellers, Christmas |
| Average word score per sentence | 37 words | 2 | Watts, six travellers |
| Simply take the first three sentences | 86 words | 3 | Watts, six travellers, fourpence |
| Any of the three methods | Varies | Never | Rochester |
Adding up word scores favours long sentences, because more words means more points, so the "summary" is 289 words of Dickens's longest and most rambling sentences, including a landlady's dialect. Averaging the scores instead swings the other way, choosing three tiny lines of dialogue that total 37 words. The plainest method of all, taking the opening three sentences, does as well as either, a simple baseline that proves hard to beat here. And none of the three keeps the word "Rochester", which appears only once in the chapter text. A method that rewards repetition cannot value a fact that is stated once, however important it is.
The learner finishes by writing the five facts a good summary must keep before looking at any output: who founded the charity, for how many travellers, what each received, where and when. Measuring every method against that list turns "this summary looks fine" into a number. That habit matters even more with AI chatbots, which write fluent new sentences rather than copying old ones. A fluent summary can still drop the place, change the number of travellers, or add a detail that was never there, and only a reader who checks against the source will notice.
Ages 8 to 11
Summarise a short story in three sentences, then tick off the key facts a friend lists.
Ages 11 to 15
Count the most frequent words in a chapter with Python and pick the top sentences.
Ages 15 and up
Compare scoring methods, fix the length bias and measure each against a fact list.
Dickens's story, our summariser
The text is Project Gutenberg's edition of The Seven Poor Travellers by Charles Dickens. The sentence splitting, the scores and every summary in the table are our own work.
