funderburkjim
commented
5 years ago feminine stem generation
The feminine stem is generated by adding a long 'I' to the citation form, but with one additional twist. For example, feminine stem for 'gAmin' is 'gAminI'.
The twist involves the need to apply nR sandhi. For instance, feminine stem of 'kArin' starts as 'kArinI', but then nR sandhi application changes this to 'kAriRI'.
As in the declension algorithms, here also we use the pada structure to determine the impact of nR
sandhi.
For instance, the feminine of agra-gAmin is taken to be agra-gAminI, since the 'r' in 'agra' is not in the final pada, and hence does not play a role.
Since the pada-structure is determined indirectly from the MW markup, there can be the possibility that a different pada structure is appropriate for our declension purposes. As example,
'akzamin' . From lexnorm-all2, this is 'a-kza-min'; and from this the feminine would be 'akzaminI', since
the 'z' is not in last pada.
But this 'a-kza-min' pada structure should be 'a-kzamin' (kzamin = patient, a-kzamin = not patient). And with this pada structure, the feminine stem is 'a-kzamiRI'.
a list to examine
To examine possible additional cases like a-kza-min, where we need to change pada structure in lexnorm-all2 to correctly determine the feminine stem, I made a list of potential questionable feminine stems for nominals endin in 'in'.
There are only 114 cases listed. These are cases where
- the computation of feminine stem based on pre-existing pada structure in lexnorm-all2 yields a feminine stem ending in 'nI'
- getting rid of all pada-structure (i.e., assuming there is just one pada) yields a feminine stem ending in 'RI'.
I hope @SergeA and/or @drdhaval2785 will take a look at this list with an eye towards finding cases like 'a-kza-min'.
I'm particularly suspicious of the cases where we have the 'in' form of a prefixed verb. For instance 'nir-vahin', 'pari-kampin', etc.
When we get more inputs on these, I'll change the pada structure in lexnorm-all2 cases for a-kza-min and similar cases.
There are 6722 nominal citations in lexnorm-all2 which end in 'in'.
The stem generation results in models m_in, n_in, and f_in_I.
See next comment for discussion of f_in_I.