Fix single letter words being treated as abbreviations
Issue #32 (closed) adds single letter abbreviation detection. This slightly improves the benchmark accuracy in some languages, while in a few (ee, fon_Latn, ht, kmr_Latn, si) the accuracy is slightly reduced.
Isaac: The reduced accuracy seems to be due to some languages having common single-character words that end sentences (but are being treated as abbreviations). Possible solutions: Allow exceptions to single-letter abbreviations and perhaps even auto-generate some based on Wikipedia occurrences.