NATURAL SELECTION 81 



advantage, but some character which is never found apart from 

 it. May not correlation, too, have lengthened the forelegs of 

 the giraffe at the same time that the neck elongated ? May it 

 not help to settle other difficult questions of co-adaptation ? It 

 must be owned, though, that it sometimes may put the drag on 

 evolution. Suppose that in birds a long neck was once corre- 

 lated with long legs, and that a particular bird required, as a 

 swan does, a long neck and short legs. But even the tie of 

 correlation may sometimes be snapped. If breeders chose, they 

 might possibly produce a breed of pigeons in which short beaks 

 were not associated with short feet. 



Meanwhile we may hold that correlation may possibly account 

 even for the varieties of grooving or for the non-grooving of the 

 stalks of buttercups. 



(3) We must now consider the bearing of isolation on this or (3) due 

 question. In a later chapter I shall try to show that without to 

 isolation favourable variations would be swamped by intercross- 

 ing, and that the building up of species could not proceed. Here 

 I must very briefly anticipate part of what I have to say there. 

 Suppose that some members of a species develop some harmless 

 peculiarity, and suppose that these by transference to a secluded 

 valley or by other means become isolated, this peculiarity will be 

 likely to survive. Inter-sterility with the parent species might soon 

 arise, so that the off-shoot species would be able to leave its seclu- 

 sion and share the habitat of the parent species without intercross- 

 ing. Once given the isolation, due to geographical separation or 

 intersterility or whatever cause, it is quite intelligible that dis- 

 tinctive characters should appear. What calls for explanation is 

 not the existence but the constancy of a small feature which can- 

 not be called an adaptation, and therefore cannot be maintained 

 by natural selection. Such features may, as I have said, be 

 maintained by correlation far oftener than we suspect. But the 

 appeal to ignorance is a dangerous argument to use often. I 

 think it is safer to take this line : nearly all the points in 

 question are comparatively insignificant : in such trifling points 

 isolation unaided seems able to produce constancy, though there 

 are no doubt slight variations in them which escape our notice as 



