fst-infl, fst-infl2, fst-infl3

fst-infl [ options ] file [ input-file [ output-file ] ]
fst-infl2 [ options ] file [ input-file [ output-file ] ]
fst-infl3 [ options ] file [ input-file [ output-file ] ]

-t file

Read an alternative transducer from file and use it if the main transducer fails to find an analysis. By iterating this option, a cascade of transducers may be tried to find an analysis.

-b

Print surface and analysis symbols. (fst-infl2 only)

-n

Print multi-character symbols without the enclosing angle brackets. (fst-infl only)

-d

The analyses are symbolically disambiguated by returning only analyses with a minimal number of morphemes. This option requires that morpheme boundaries are marked with the tag <X>. If no <X> tag is found in the analysis string, then the program (basically) counts the number of multi-character symbols consisting entirely of upper-case characters and uses this count for disambiguation. The latter heuristic was developed for the German SMOR morphology. (This option is only available with fst-infl2 and fst-infl3.)

-e n

If no regular analysis is found, do robust matching and print analyses with up to n edit errors. The set of edit operations currently includes replacement, insertion and deletion. Each operation has currently a fixed error weight of 1. (fst-infl2 only)

-% f

Disambiguates the analyses statistically and prints the most likely analyses with at least f % of the total probability mass of the analyses. The transducer weights are read from a file obtained by appending .prob to the name of the transducer file. The weight files are created with fst-train. (fst-infl2 only)

-p

Print the probability of each analysis. (fst-infl2 only)

-c

use this option if the transducer was compiled on a computer with a different endianness. If you have a transducer which was compiled on a Sparc computer and you want to use it on a Pentium, you need to use this option. (fst-infl2 only)

-q

Suppress status messages.

-h

Print usage information.

fst-infl is a morphological analyser. The first argument is the name of a file which was generated by fst-compiler. The second argument is the name of the input file. The third argument is the output file. If the third argument is missing, output is directed to stdout. If the second argument is missing, as well, input is read from stdin.

fst-infl2 is similar to fst-infl but needs a transducer in compact format (see the man pages for fst-compiler and fst-compact). fst-infl2 is implemented differently from fst-infl and usually much faster.

fst-infl3 is also similar to fst-infl but needs a transducer in lowmem format (see the man pages for fst-compiler and fst-lowmem). fst-infl3 accesses the transducer on disc rather than reading it into memory. It starts very fast and needs very little memory, but is slower than fst-infl2.

fst-infl reads the transducer which is stored in the argument file. Then it reads the input file line by line. Each line is analysed with the transducer and all resulting analyses are printed (see also the man pages for fst-mor).

No bugs are known so far.

fst-compiler, fst-mor

Helmut Schmid, Institute for Computational Linguistics, University of Stuttgart, Email: schmid@ims.uni-stuttgart.de, This software is available under the GNU Public License.