The filter is in "ipa" -- it should be placed before any other
troff preprocessors that it interacts with (e.g. tbl).

An example of its use can be found in test_input, which is
processed via 

	ipa test_input | tbl | troff -mm | dcan


The source for the crucial part of the filter is ipa_trans.lex --
it would be easy to modify the ascii-to-ipa mapping (e.g. for arpabet,
or for a different language).

The command "ipa_key" will print a copy of the font map and
of the ascii-to-ipa mapping designed for English.

Notice that other use of font positions 5 and 6 will tend not to work.
There isn't any easy and general way around this problem, it seems.

At the moment, "ipa_trans" does not check that a final delimiter
is ever found.

The ipa fonts are to found in the usual places (e.g. /usr/can/font,
/usr/jerq/font, /usr/lib/font/dev202).
