Search Descriptions

Main Topics

Search Publications


author

title

other

year

Transliteration with Finite State Machines

Since transliteration is a monotone process without reordering, finite state machines have been used in early work.

Transliteration With FSM is the main subject of 9 publications.

Publications

Knight and Graehl (1997) present a model that maps between letters and phoneme representations. Such models may be extended by using a larger Markov window during mappings, i.e. using a larger context (Jung et al., 2000). Schafer (2006) compares a number of different finite state transducer architectures. For closely related language pairs, such as Hindi–Urdu, deterministic finite state machines may suffice (Malik et al., 2008).
Transliteration may use either phonetic representation to match characters of different writing systems (Knight and Graehl, 1997) or map characters directly (Zhang et al., 2004). Phoneme and grapheme information may be combined (Bilac and Tanaka, 2004). Given small training corpora, using phonetic representations may be more robust (Yoon et al., 2007).

Benchmarks

Discussion

Related Topics

New Publications

  • Knight (2009)

Actions

Download

Contribute