Two-Level Phonology/Morphology

• Modeling a phonological rule with a finite-state machine
  • FS transducer (corresponding to regular relation) associating lexical inputs with surface outputs
• FST works for both production and recognition
• No ordering or intermediation representations are required; FSTs for each rule operate in parallel.

  

• Kinds of rules
  • 

    A lexical t or a surface c before an i correspond to one another.

  • 

    If a lexical t corresponds to a surface c, it precedes an i.

  • t:c <=> ___ i
    A lexical t corresponding to a surface c always precedes an i.
  • t:c <\= ___ i:e
    If a lexical i corresponds to a surface e, then a lexical t before it cannot correspond to a surface c (or vice versa).
• Example: Japanese past tense
  • Voicing: t:d <=> <b m n g>: (+:0) (0:i) ___

    

  • Epenthesis: 0:i <=> <s k g>: (+:0) ___ t:

    

  • Velar deletion: <k g>:0 <=> ___ (+:0) 0:i t:

    

[L645 Syllabus]