Finite State Transducers

1
Finite State Transducers

• The machine model we will study for morphological
  parsing is called the finite state transducer
  (FST)
• An FST has two tapes
• input tape (with an input alphabet)
• output tape (with an output alphabet)

2
Formal definition of FST (from text)

• M (Q, å, q0, F, d), where
• Q is a finite set of states
• å is a finite alphabet of complex symbols (i.e.
  pairs of input-output symbols). å io i is
  an input tape symbol and o is an output tape
  symbol
• q0 Î Q is the initial state
• F Í Q is a set of final (accepting) states
• d Í Q å

3
Example

• We want to be able to parse words (recover
  structure for them) including such words as goose
  (which is ambiguous)
• goose ? goose N SG or goose V
• geese ? goose N PL
• gooses ? goose V 3SG

4
Components of a morphological parser

• lexicon morphemes (stems and affixes) together
  with category information
• morphotactics rules of morpheme order
• orthographic (spelling) rules rules of changes
  in spelling when morphemes combine

5
Lexicon for FST

• The lexicon can be modelled using two levels
• Surface form (e.g. geese)
• Underlying form (e.g. goose N PL)
• This will allow lexicon to handle irregular forms
• Example lexicon on next slide

6
Example lexicon
7
Generation example foxes
8
FST for fox N PL ? foxs
ff
oo
xx
Ne
PLs
q0
q1
q2
q5
q6
q7
9
FST for E-insertion rule
other
other
z,x

?e
s
q0
q1
q2
q3
q4
? other
?
z,x,s
z,x,s
?
s
z,x,s
q5
other
other means any symbol except s, x, z,
, ?,
10
Generation example foxes
0
1
2
5
6
7
2