Suomilog is a toolkit for parsing context-free grammars that have embedded morphological information. It's intended use is parsing Finnish sentences and it includes a module that can parse and generate inflected Finnish sentences using libvoikko as back-end.

The suomilog module is used for parsing context-free grammars. Suomilog grammars are context-free grammars that have additional morphological information. Below is an example.

.FEATURE ::= .HUMAN{+gen} nimi{$}
.FEATURE ::= .HUMAN{+gen} ikä{$}

.HUMAN ::= mies{$}
.HUMAN ::= nainen{$}
.HUMAN ::= ihminen{$}
.HUMAN ::= .ADJECTIVE{$} .HUMAN{$}

.ADJECTIVE ::= ahkera{$}
.ADJECTIVE ::= kaunis{$}
.ADJECTIVE ::= erittäin .ADJECTIVE{$}

In Suomilog grammars nonterminals are marked with a dot and all other symbols are terminals.

Morphological information is written in braces after a symbol. For example kaunis{+gen} would match the word kauniin. A dollar $ means that the morphological information is passed forward. So .HUMAN{+gen} for example mathes miehen, naisen, ihmisen, ahkeran miehen, and so on.

The braces can contain multiple form names separated with commas. For example, tehdä{+inf3,+gen} would match tekemisen. In code these form names are called "bits".

For example usage, see the examples/ folder.

The suomilog.finnish module contains tools for Finnish morphological parsing and generation. It uses pypykko as its back-end.

The function suomilog.finnish.tokenize(text) is used to tokenize words:

import suomilog.finnish as f
f.tokenize("kissa käveli kadulla")
# outputs:
[
    Token('kissa', [('kissa', {'', '+sg+nom', ':noun', '«kissa»', 'kissa:noun', 'kissa:', '+sg', '+nom'})]),
    Token('käveli', [('kävellä', {'', '+3sg', '«käveli»', 'kävellä:', ':verb', 'kävellä:verb', '+past'})]),
    Token('kadulla', [('katu', {'', ':noun', 'katu:', '«kadulla»', '+ade', '+sg', 'katu:noun'})])
]

The function suomilog.finnish.inflect_nominal(word, plural, case) is used to inflect nouns, adjectives and numerals:

import suomilog.finnish as f
print(f.inflect_nominal("kissa", "+pl", "+par")) # kissoja