I went through ECH's document, Moses Feature System.odt, and noted several cases where we may need to further refine the feature system.

Is it our goal that every affix (well, besides lexical suffixes and lexical prefixes) have a unique feature structure? If so, we will need additional features to disambiguate the following:

- the applicative suffixes ɬ-DIR, tuɬ, and xit. (MLW 2003 could not identify differences in usage for these sufffixes.)

- ɬ-DIR and ɬ-EP can be distinguished with the symbol values "applicative" and "external-possession", respectively, but their meanings overlap somewhat. They are currently combined into one entry, "ɬt" in the database.

- all locative prefixes (kat-LOC, k-LOC, kɬ-LOC, kʼɬ-LOC, n-LOC, niʔ, t-LOC)

- the directional prefixes (ʔal, c, lc)

- the nominalizer-instrumentals min-INST and tn.

- the "count" prefixes: kʼɬ-DER (for counting sacks or bags), kɬ-DER (for ordinal numbers), and k (for counting people)

- the irrealis mood prefixes: kas for verb stems, and kaɬ-PR for noun stems. (Should these be allomorphs within one entry? Or two entries, cross-referenced?)