Can realism be retro-fitted?

I've been wondering about the possibility of taking some of my existing
languages and developing their historical background to make them more
realistic and less artificial-seeming. I recently finished reading Guy
Deutscher's _The Unfolding of Language_ which has been mentioned here
lately; one of the things that stood out was the role of analogy in
creating new patterns. I'm familiar in a general sense with how the
sounds and meanings of words can change over time. But one of the
problems I've always had in reconstructing the history of a language
that doesn't have one is trying to come up with a consistent set of
rules that relate the sounds of the hypothetical older language to the
already documented language.

The problem with taking an existing language to start with is that I
have to come up with a historical explanation for each feature of the
language, or modify it in such a way that I can more easily explain it.
For instance, when I examined the tone patterns of two-syllable words in
Simîk, I noticed that a few patterns were much more comman than the
others, which could be explained by development from a simpler tone
systme in earlier versions of the language. But not all tone patterns
fit into that system, so I had to assume they were borrowed from some
other language, or were different in some other way (having a special
tone pattern that was used for emphasis).

In the long run, is it better to start with one or more artificial
proto-languages and develop them forward through time (which involves a
lot of work on features that may not even make it into the future
language system), or to start with an existing language and develop a
history for it?

For a specific example, I thought of taking Tirelat and trying to
develop a history for it. Tirelat is a very regular and artificial
language, which may actually be a result of engineering a more natural
language to eliminate irregularities. But to start with something
simple, the vowel system: Tirelat has 7 vowels /a e i @ 1 o u/, which
may be long or short. Diphthongs do exist, but only /ai/ is common;
combinations like /ia/ and /ui/ can be analyzed as consonant + vowel
(/ja/, /wi/), except for the fact that there is no /ji/ or /wu/. So
where do these 7 vowels come from?

One thought that might explain the long vowels is that earlier versions
of the language had more diphthongs, which simplified to single vowels.
Notably, /a i u/ are more common than the other vowels. But any patterns
in the data would have to be coincidental at this point, since the
Tirelat vowel system wasn't developed with a history in mind. So in a
case like this, would it be better to come up with an arbitrary history
that doesn't fit all the facts (e.g. proto-language /o/ develops into
/@/ except in the vicinity of a bilabial consonant, where it remains
/o/, after which other phonetic changes occurred which caused /f/ to
merge with /x/ resulting in a phonemic distinction between /x@/ and
/xo/)? As it happens, /xo/ is slightly more common than might be
expected, but /p@/ and /b@/ do exist, which can't be explained by the
hypothetical sound changes.

Replies