tical, excepting their respective orthography. Such a system must necessarily quantify a
text based on quantified semantic value, rather than morphological value, for, as we have
just shown, morphology is not necessarily consistently represented between two parallel
copies.
There is another alternative open to us that still allows for the use of morphemes as a
quantifying unit of calculation. That method would be to reconstruct a cuneiform text’s
morphology based on what the logograms imply, thus equating logographic forms to syl-
labic forms in the statistical analysis. However, this solution creates a greater methodo-
logical problem than it resolves: it generates statistical evidence based on conjectural
readings that find no graphical representation in the texts themselves. By employing such
a methodology we would have to concede that some of the data included in our analysis
would come from morphological material that was essentially invented by the process of
examination itself. Such an analysis could not claim to accurately represent the texts un-
der examination.
These considerations raise the question as to which morphemes should be ignored to
avoid conjectural values leaking into the statistical results. Such cases as unwritten defi-
nite forms (definite articles assimilated to prepositions, or in construct chains), and so-
called ‘zero’ morphemes (the 3ms suffix conjugation verb) should be ignored because
they are not represented graphically. In addition to these non-graphic morphemes we
should exclude morphemes that are semantically bound to other morphemes. We are
thinking here in particular of plural markers and markers of gender, which are semanti-