Normatex - Russian text normalization

This is a set of Finite-State Transducers (FSTs) for normalization of Russian texts for speech synthesis, machine translation and other natural language processing tasks.

The FSTs are developed using Unitex, a corpus processor.

To normalize a Russian text:

1. Copy your text (e.g. example.txt) to Corpus folder, open it in Unitex and preprocess it with following resources:

• apply Graphs/Preprocessing/Sentence/SentenceUniver.grf in MERGE mode
• apply Graphs/Preprocessing/Replace/replace.grf in REPLACE mode

2. Apply lexical resources:

• the full version of the Russian computational morphological dictionary developed at CIS, Munich: CISLEXru.bin, CISLEXru_disamb-.bin and CISLEXru_EN.bin
• Dela/univer.bin
• Dela/univer_disamb-.bin

3. Create a cascade (Text\Apply CasSys Cascade... menu, New) to sequentially apply the following FSTs to your text in REPLACE mode:

• Graphs/numbers.fst2
• Graphs/abbr/abbr_w.fst2
• Graphs/abbr/acronyms_w.fst2
• Graphs/Postprocessing/replace.fst2

4. Launch the cascade of FSTs.
5. The normalized text is in Corpus/example_csc/example_4_0.snt.

Slides