Word study

forever

Two movements shape the record: first the spelling balance turns as for ever retreats; later both forms rebound after the split has already settled.

01 / LONG ARC

Visibility / form

exact-form appearances per million corpus word tokens

68.5959.489.11
37.2718.6818.59
32.8114.2418.57
12.222.699.53
35.656.8928.75
forever for ever
Selected decade exact-form rates per million corpus word tokens
Decadeforeverfor everpair total
1820s9.11361613959.47736392868.590980067
1880s18.58601481118.68186175137.267876562
1890s18.57222874414.2392439632.811472704
1980s9.5313999472.68999015412.221390101
2010s28.7534604356.89206432535.64552476

PAIR CURRENT

Peak / low / return

68.591820s / pair peak
12.221980s / pair low
35.652010s / return
Peak→low
-82.18%
Low→2010s
2.92×
Below peak
-48.03%

A partial return in one fixed corpus, not a full recovery.

DATA CARDS

Peak / low / return

1 / 3

1820s / PEAK68.59exact-form appearances per million corpus word tokensDETAIL / SOURCE +
Definition
The highest selected long-arc anchor.
Reading
The pair is unusually visible in the 1820s, led by the spaced form.
Boundary
This is a fixed-corpus decade aggregate, not a language-wide peak.
Source
googlebooks-eng-20200217; 1820–1829; exact raw rows and same-release totalcounts-1.
1980s / LOW12.22exact-form appearances per million corpus word tokensDETAIL / SOURCE +
Definition
The lowest selected long-arc anchor.
Reading
Both exact forms reach the pair's selected low before the later return.
Boundary
A low in this corpus does not establish absence from language.
Source
googlebooks-eng-20200217; 1980–1989; exact raw rows and same-release totalcounts-1.
2010s / RETURN35.65exact-form appearances per million corpus word tokensDETAIL / SOURCE +
Definition
The latest complete decade in the fixed release.
Reading
The pair returns to 2.92 times its 1980s rate but remains below the 1820s anchor.
Boundary
The 2010s are a partial return, not a full recovery.
Source
googlebooks-eng-20200217; 2010–2019; exact raw rows and same-release totalcounts-1.

02 / FIRST TURN

Majority without growth

The joined form becomes the decade-level majority while its own rate remains effectively flat.

ANNUAL SEQUENCE1884
55%50%45%
188218831884188518861887

1884 / the first crossing in this six-year sequence.

CHANGE LEDGER
188218831884188518861887
forever
-0.07%
for ever
-23.78%
pair total
-11.96%

First sustained reach lead / 1886

DATA CARDS

The annual turn

1 / 3

1884 / FIRST CROSSING50.16%joined share of pair exact-form appearancesDETAIL / SOURCE +
Definition
Joined exact-form appearances divided by joined plus spaced exact-form appearances in the same year.
Reading
The first annual crossing in the inspected six-year sequence.
Boundary
A yearly crossing is not an irreversible language event; the fixed corpus and its composition bound the result.
Source
googlebooks-eng-20200217; 1884; exact raw rows and same-release totalcounts-1.
1885 / REVERSAL46.33%joined share of pair exact-form appearancesDETAIL / SOURCE +
Definition
Joined exact-form appearances divided by joined plus spaced exact-form appearances in the same year.
Reading
The following year falls back below equal share.
Boundary
A yearly crossing is not an irreversible language event; the fixed corpus and its composition bound the result.
Source
googlebooks-eng-20200217; 1885; exact raw rows and same-release totalcounts-1.
1886 / SUSTAINED53.82%joined share of pair exact-form appearancesDETAIL / SOURCE +
Definition
Joined exact-form appearances divided by joined plus spaced exact-form appearances in the same year.
Reading
The joined form crosses again and remains above in 1887.
Boundary
A yearly crossing is not an irreversible language event; the fixed corpus and its composition bound the result.
Source
googlebooks-eng-20200217; 1886; exact raw rows and same-release totalcounts-1.

03 / SECOND TURN

Rebound without another switch

Both forms return after the spelling balance has already settled near four-fifths joined.

1980s pair / total 100%

2010s pair / total 100%

1.38 pp1990s–2010s joined-share band width / 80.66–82.05%

Share is calculated only within the two exact forms; it is not language-wide spelling adoption.

DATA CARDS

What drives the return

1 / 3

forever / JOINED RETURN3.02×1980s→2010s exact-form rate factorDETAIL / SOURCE +
Definition
Rate factor decomposed into reach × repetition ÷ corpus words-per-volume factor.
Reading
Reach 1.938×; repetition 1.306×; corpus density ÷0.839×.
Boundary
This decomposition is an accounting identity, not a causal attribution.
Source
googlebooks-eng-20200217; complete 1980s and 2010s raw rows and totalcounts-1.
for ever / SPACED RETURN2.56×1980s→2010s exact-form rate factorDETAIL / SOURCE +
Definition
Rate factor decomposed into reach × repetition ÷ corpus words-per-volume factor.
Reading
Reach 1.352×; repetition 1.589×; corpus density ÷0.839×.
Boundary
This decomposition is an accounting identity, not a causal attribution.
Source
googlebooks-eng-20200217; complete 1980s and 2010s raw rows and totalcounts-1.
pair / COMBINED RETURN2.92×1980s→2010s combined exact-form rate factorDETAIL / SOURCE +
Definition
Combined joined plus spaced exact-form rate in two complete decades.
Reading
12.22 in the 1980s → 35.65 in the 2010s.
Boundary
The factor describes this fixed corpus release; it is not a causal attribution.
Source
googlebooks-eng-20200217; complete 1980s and 2010s raw rows and totalcounts-1.

04 / CONDITIONS

Breadth or repetition?

Switch between three valid conditions. The paired decade marks redraw on each condition’s own scale.

CONDITION COMPARISON4.172×

In the 2010s, joined visibility is 4.172 times the spaced exact form.

In the 2010s, forever appears in 51.84% of corpus volumes and for ever in 13.06%; the joined form therefore reaches 3.970 times as many volumes.

In the 2010s, the joined form is repeated only 1.051 times as often inside a containing volume.

forever for ever
forever / 2010s
28.75
for ever / 2010s
6.89

exact-form appearances per million corpus word tokens

CLOSING FINDING

Found in more books, not repeated much more.

By the 2010s, forever appears in many more books in this Google Books sample than for ever. But in books where either spelling appears, the two are repeated at almost the same rate.

That means forever’s modern lead comes mainly from reaching more books, not from being used over and over within the same book.

Three measures
Overall use
4.172×

In the 2010s sample, forever appears about four times as often per million words as for ever.

Use within a book
1.051×

In books that contain the spelling, forever is repeated only about 5% more often than for ever.

How steady the split stays
1.38 points

From the 1990s through the 2010s, forever stays close to four-fifths of the two spellings’ recorded appearances. Its share moves by only 1.38 percentage points.

These figures apply only to forever and for ever in this fixed Google Books sample. They do not represent all English writing, identify when a spelling became accepted, or show that one meaning replaced another.

Citation / rights / links

Rates and volume conditions use the fixed Google Books English 2019 release, persistent identity googlebooks-eng-20200217, for complete decades from 1800–2019.

The fixed corpus reflects the books Google retained and processed; OCR error, metadata error and changing corpus composition remain. Exact surface forms do not identify meaning, first use, social acceptance or language-wide spelling adoption. Sparse raw absence is absent_or_suppressed and is never silently converted to observed zero. Viewer-normalised fractions are not used for the shared joined/spaced scale.

Google Books Ngram data remain subject to Google's source conditions. Project code is available under its repository licence; original research, writing and design remain credited to Dai Pan / 潘岱.

Back to top