The last lesson stated rules. This one counts them. Each rule is run over 1,210 regular verbs and checked against the forms a word list records, and each is right more than 99 times in 100. The nine words the past rule misses are the content: every one has a reason, and one rule added for good reasons made the score worse.
Core ideas
Three ideas, and the last is the one that saves time.
The lab runs the rule on every verb in the printed list and marks it right when the form it writes is one the list records. The share right is the number of rights over the number of verbs. Nobody had to be asked their opinion, and you can read every row.
A rule that is right 99.26% of the time is wrong on nine verbs. Those nine are not random. A British spelling, a letter no rule knows about, a word that doubles against the strong part: each kind can be named, and a named kind is something you can learn once.
The alternative to a rule is a list of every form of every verb. The rule is right on all but nine, and the nine fit on one line. That is the whole saving, and it is why the rule is worth learning.
The material
The rules are run over every regular verb in a list of the 2,800 most common words, and checked against the forms that list itself records. Here is how they did, and what they missed.
A rule with no number beside it is only an opinion. The last lesson gave you four rules that build the forms of a verb. This one asks how good they are, and the answer is a count.
The rules are run, on this page, over the 1,210 regular verbs in a list of the 2,800 most common English words. For each verb the rule writes a form, and the form is checked against the forms the list already records for that verb. A match is a hit and anything else is a miss. The lab prints both.
They are right 99.92% of the time for -s, 99.34% for -ing, and 99.26% for the past. All three are counted on this page: 1209 of 1210, 1202 of 1210 and 1201 of 1210. The words they miss are listed in the lab and again in the worked example, and there are not many.
The nine words the past rule misses are not random, and not all of them are mistakes. They sort into five kinds.
Each kind has a reason you can state. That is what a hit rate should come with, and it is why the lab prints the words and not only the number.
The word list was made by a machine, and a machine can record a spelling nobody writes. The raw list had offerring, commiting and councilling; it gave verb forms to words that are not verbs, such as able and son; and it recorded comed and maked, so the past rule was being marked right for forms nobody writes. Scored against that list, a right rule looks wrong and a wrong rule looks right.
So the list was cleaned by one test, applied to every spelling the same way: a spelling stays if a dictionary, a book that records the real spellings of a language, has it, or if it is the British spelling of one the dictionary has. In all, 146 words were left out, each with its reason, and the lab prints them all under List. Ninety of them are irregular verbs, left out because their past is not made by any rule; the course called Irregular Verbs takes them.
When the -s rule first ran, every -o took -es, and it turned radio and video into forms nobody writes. Choose the -s rule before the -o fix in the lab and you can watch it: 1207 of 1210, 99.75%, with radio and video among the misses. The fix was easy: -o takes -es only after a consonant, as in goes, and plain -s after a vowel.
At the same time a second rule was added, because it also looked right: -f becomes -ves. For nouns that is often true: one leaf, two leaves. Choose it in the lab and the count falls to 1205 of 1210, 99.59%, because brief, golf, proof and roof are verbs and they simply take -s.
This is worth more than any single rule in the lesson. A rule can sound right, come from a real pattern, and still cost you accuracy. The only way to know is to run it over real words and count. That is what every lesson on this path does.
Interactive
Every cell is formed on the page by the rule named beside it
The rules scored on every regular verb in the printed list below.
| time | simple | progressive | perfect | perfect progressive |
|---|
Method
To judge any rule, do this.
Pick one under Score a rule on the list. The lab names it in words beside the score.
Right is how many verbs the rule got right out of how many it was run on. Share right is the same count as a percent.
The table under the score lists each one, with the form the rule wrote and the forms the list records.
A spelling that differs by country, a letter the rule does not know, a strong part it cannot hear. A kind with one word is a word to learn; a kind with ten is a new rule.
Choose the past rule in the lab. It is right on 1201 of 1210 verbs, and the table lists the nine it is not.
Initial and counsel are written one way in British English and another in American English. The list records only the British -ll-, so the rule's American form is marked wrong where nothing is wrong. Panic and traffic take a -k before the ending, panicking, which no rule here knows. Format, input and output double their last letter though the strong part comes first. Bus is busing in the list where the rule writes bussing, and writers use both. And up has no consonant before its vowel, so the rule never doubles it.
The -ing rule misses the same words without counsel, which is eight. The -s rule misses one: stomach ends in -ch but does not hiss, so the rule writes stomaches and the list records stomachs. The rule reads letters and cannot hear.
You should be able to run each of the three forming rules over the list, read its score and every word it misses, sort the misses into kinds, say which rule change made the results worse and why, and explain why a list can make a right rule look wrong.
Something to change about this lesson? Write it down, tick it off when it is handled, and export the list when you want to act on it. Kept in this browser only: nothing is sent anywhere, and nobody is notified.
No recommendations yet for this lesson.