An FST that generates Swahili verbs from a segmented representation
Morphotactics (2)
Dependencies within the word
FSTs that analyze and generate ungrammatical words
Non-concatenative morphology
Circumfixes: prefix and suffix combinations that function as a unit
Root + pattern morphology; a Tigrinya example
Reduplicationninasemasema 'I speak repeatedly'
Alternation rules
Rules governing variation in the form of morphemes in particular contexts
An English example y → ie / C __ +s
(A context-sensitive rule stating that y turns into ie when it comes after a consonant and before a morpheme boundary followed by
s.)
An English orthographic rule implemented as an FST
+ → e / S _ s # S = {s, z, h, x}
Interaction among rules; the order matters
Two Oromo rules
0 → i / CC_+
t → d / B+_; B = {b,d,g}
In the analysis argita → arg + ta,
the first rule has to precede the second
A single FST for a word class
FSTs can be composed, that is,
given a transducer T on alphabets Σ and Γ and a transducer S on alphabets Γ and Δ, there exists a transducer T ⊙ S on Σ and Δ such that x[T ⊙ S]z if and only if there exists a string y ∈ Γ* such that x[T]y and y[S]z.
To create a single FST for a word class (or all words) in a language, we compose
The FSTs for all alternation rules, in the appropriate order
An FST for the morphotactics, produced by concatenating
the FSTs for each morpheme class and the lexicon of roots (or stems)
A "guesser" analyzer has no lexicon of roots/stems but instead an FST that
encodes the possible forms that roots/stems take; it permits
analysis of words based on novel roots/stems. muzzes ↔ muzz+s
