Creation of Spoken Hebrew Databases

What it is about

A company (NSC) describes two telephone speech databases of Modern Hebrew for speech recognition in teleservices, designed after the SpeechDat fixed-network model: DB I with 250 speakers × 32 utterances and DB II with 850 speakers × 75 utterances, read and spontaneous. Speakers were sampled by age, gender, education and, instead of region, by country of origin (Ashkenazi and Sephardi pronunciations; the Sephardi one keeps guttural allophones). Read prompts were printed with full niqqud to remove ambiguity (the paper's example: ספר as sefer "book", sapar "hairdresser", safar "counted"). The material covers digits in both genders, digit strings, numbers, money, times and dates (from BNF grammars of possible phrases), spelled letters, about 70 application keywords given in masculine and feminine forms (spoken Hebrew uses the future for the imperative), 600 phonetically rich sentences and about 1,000 phonetically rich words with common foreign words. Everything was transcribed orthographically and in SAMPA. The automatic text extraction of this PDF failed (font-encoded Hebrew), so this card was written from the page images of pages 1–3.

Ideas

• I1. Number, time and date phrases for Modern Hebrew as small BNF grammars (gendered digits, units, "quarter past", relative dates), as the paper did for its prompts. A Rust grammar of this kind is a ready tokenizer/normaliser stage for numerals in Modern Hebrew text, and a generator of test sentences.
• I2. The future-for-imperative usage of spoken Hebrew (the paper's reason for listing keywords in the future) is a register note and a rule for the Modern Hebrew verb chapter: an imperative frame may be expressed by a 2nd-person future.
• I3. The sefer/sapar/safar triple is a minimal test of unpointed ambiguity; collect such homograph sets as the first test for any Modern Hebrew reader.

For Mova

I1 goes into 2. Script and spelling (numbers and dates) and the plan in 12. Mova and Modern Hebrew: where it stands, what it would need; I2 into 05. The verb: system, binyanim, tense and use and 09-pragmatics-politeness.md; I3 into 2. Script and spelling. The databases themselves are commercial and not available to Mova.

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{rannon2000creation,
  title = {{Creation of Spoken Hebrew Databases}},
  author = {Rannon, Tami and Golani, Ofra and Goren, Anat and Shammass, Sherrie and Moyal, Ami},
  booktitle = {Proceedings of the Second International Conference on Language Resources and Evaluation (LREC'00)},
  year = {2000},
  url = {https://aclanthology.org/L00-1037/}
}

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lingbulk-2000-rannon-creation-spoken-hebrew-databases/