1 / 24

FST Morphology

FST Morphology. Miriam Butt October 2002. Based on Beesley and Karttunen 2002. Recap. Last Time: Finite State Automata can model most anything that involes a finite amount of states. We modeled a Coke Machine and saw that it could also be thought of as defining a language.

brooke-levy
Download Presentation

FST Morphology

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. FST Morphology Miriam Butt October 2002 Based on Beesley and Karttunen 2002

  2. Recap Last Time: Finite State Automata can model most anything that involes a finite amount of states. We modeled a Coke Machine and saw that it could also be thought of as defining a language. We will now look at the extension to natural language more closely.

  3. A One-Word Language c a n t o

  4. A Three-Word Language i g r e t c a n t o m a e s

  5. Analysis: A Successful Match i g r e t c a n t o m a e s Input: m e s a

  6. Rejects The analysis of libro, tigra, cant, mesas will fail. Why?

  7. Transducers: Beyond Accept and Reject i g r e t i g r e t c a n t o c a n t o m a e s m a e s

  8. Transducers: Beyond Accept and Reject Analysis Process: • Start at the Start State • Match the input symbols of string against the lower-side symbol on the arcs, consuming the input symbols and finding a path to a final state. • If successful, return the string of upper-side symbols on the path as the result. • If unsucessful, return nothing.

  9. A Two-Level Transducer i g r e t i g r e t c a n t o c a n t o m a e s m a e s Input: m e s a Output: m e s a

  10. A Lexical Transducer +PresInd +Sg +Verb c a n t a r +1P c a n t 0 0 0 0 o 0 Input: c a n t o Output: c a n t a r +Verb +PresInd +1P +Sg One Possible Path through the Network

  11. A Lexical Transducer c a n t o +Sg +Noun +Masc c a n t o 0 0 0 Input: c a n t o Output: c a n t o +Noun +Masc +Sg Another Possible Path through the Network

  12. The Tags Tags or Symbols like +Noun or +Verb are arbitrary: the naming convention is determined by the (computational) linguist and depends on the larger picture (type of theory/type of application). One very successful tagging/naming convention is the Penn Treebank Tag Set

  13. The Tags What kind of Tags might be useful?

  14. Generation vs. Analysis The same finite state transducers we have been using for the analysis of a given surface string can also be used in reverse: for generation. The XRCE people think of analysis as lookup, of generation of lookdown.

  15. Generation --- Lookdown +PresInd +Sg +Verb c a n t a r +1P c a n t 0 0 0 0 o 0 Input: c a n t a r +Verb +PresInd +1P +Sg Output: c a n t o

  16. Generation --- Lookdown Analysis Process: • Start at the Start State and the beginning of the input string • Match the input symbols of string against the upper-side symbols on the arcs, consuming the input symbols and finding a path to a final state. • If successful, return the string of lower-side symbols on the path as the result. • If generation is unsucessful, return nothing.

  17. Concatenation One can also concatenate two existing languages (finite state networks with one another to build up new words productively/dynamically. This works nicely, but one has to write extra rules to avoid things like: *trys, *tryed, though trying is okay.

  18. Concatenation w o r k Network for the Language {“work”} s n g i d e Network for the Language {“s”, “ed”, “ing”}

  19. Concatenation s k w o r g n i e d Concatenation of the two networks What strings/language does this result in?

  20. Composition Composition is an operation on two relations. Composition of the two relations <x,y> and <y,z> yields <x, z> Example: <“cat”, “chat”> with <“chat”, “Katze”> gives <“cat”, “Katze”>

  21. Composition c a t c h a t c h a t K a t z e

  22. Composition c a t c h a t K a t z e Merging the two networks

  23. Composition c a t K a t z e The Composition of the Networks What is this reminiscent of?

  24. Other Uses for the Transducers c a t T C A Upper/Lower Casing

More Related