HFST - The Helsinki Finite-State Transducer technology is intended for creating and manipulating weighted or unweighted synchronic transducers implementing regular relations. UTF-8 is chosen as the character encoding used in the HFST software. Currently, HFST has been implemented using the SFST, OpenFst and foma software libraries. Other versions may be added in some future release. SFST and foma implementations are unweighted and OpenFst implementation is weighted.

Some of the HFST interface functionalities have also been implemented for HFST's own two transducer formats, HfstTransitionGraph and optimized lookup format. The previous is useful for converting between transducer formats and storing transducers in an implementation-independent format. The latter is used for fast lookup of strings in a transducer.

All implementations work according to the same interface, so it is possible to compile the same piece of code using different backend libraries. It is also possible to add your own implementation under the HFST interface.

For a quick start to the HFST interface with examples, see here.

The examples given in this documentation use Xerox transducer notation.

What is new in version 3

A third transducer library, foma, is included
Adding your own implementation under the HFST interface is easy
Unknown and identity symbols are supported

Features

Create transducers and apply operations on them
Create transducers from scratch
Iterate through a transducer's states and transitions
Create transducers by tokenizing UTF-8 strings with multicharacter symbols
Apply two-level, replace, restriction and coercion rules

Tutorial

A quick start to the HFST interface with examples
Adding your own implementation of HFST
Differences between HFST versions 2 and 3
Using HFST with SFST, OpenFst and foma

Download

Download and install the HFST interface and command line tools

Links

Examples of programs implemented using the HFST interface
Documentation of the HFST project