How can I transform the following dataset to a more linear representation? R code:
x <- 1:20
y <- c(101, 84, 81, 80, 73, 69, 67, 63, 63, 62, 62, 61, 61, 60, 59, 58, 57, 57, 57, 55)
d <- data.frame(x, y)
ggplot(d, aes(x, y)) + geom_point()
Here is a log version, almost no change:
ggplot(d, aes(x, log(y))) + geom_point()
That is, I have several datasets that look somewhat exponential, like this example, but taking a simple log isn't good enough to produce the level of linearity I need. What can I do to get more linearity?

