John Vertical wrote:

> There's a third choice too, sort of. I've found it useful to start at 
> least sketching history alreddy in the erly phases of a project; there's 
> the risk of getting sidetracked, but even a little preliminary work can 
> prevent having to do major retcons. I still primarily work backwards, 
> however, since 1) I like puzzles, and 2) I usually have a specific kind 
> of sound in mind for the present day language, and it's easier to 
> achieve it that way. So I'm not really starting from a proto-language, 
> but not really adding history to an existing language either.
> I'm afraid this'll be of little use in solving your current problems, 
> however.

Right, the early stages of the Tirelat project were back in 1999, and I 
hadn't intended it to háve a history. :-)

Admittedly, the puzzle aspects of this could be interesting to work out. 
I wonder if I can make sense of the fused time/evidential morphemes from 
some historical sequence of sound changes plus changes in meaning (with 
perhaps some borrowing from unrelated languages to fill any gaps in the 
system). On the other hand, any sound changes should have affected the 
rest of the morphology, which doesn't show any signs of it.

But hopefully at least some of my recent projects -- I've already 
started looking into the history of Yasaro, for instance -- will be 
easier to work out. So I can just decide that /ki/ changed to /tSi/ at a 
particular point in history, and I don't already have a ton of /ki/ 
words in the existing vocabulary to account for. The few I have can be 
explained by other means, such as unstressed /e/ raising to /i/, which 
took place after the /k/ > /tS/ change. And even from the start of the 
Yasaro project, I'd already decided to use a historical explanation for 
tones similiar to what I've read about from the history of (the former)