A Morphologically Annotated Hebrew CHILDES Corpus

What it is about

The paper describes the Hebrew part of the CHILDES child-language database: about 418,000 word tokens (about 13,800 types) of child-adult speech from two longitudinal corpora. The transcription is a Latin-based phonemic one that keeps the five vowels of Modern Hebrew, primary stress, and the prefix particles as separate morphemes, so that it has fewer homographs than the standard script. A morphological grammar in the CLAN MOR format has a lexicon of about 5,800 entries in 16 part-of-speech classes, over 1,000 allomorph rules and over 100 affixation rules. On a new 10,070-token corpus, 1.75 percent of tokens (5.3 percent of types) get no analysis, and an independent check against the MILA analyser disagrees on 11 percent of forms. Context disambiguation was trained on 18 of 304 hand-tagged files, with 80 percent accuracy on ambiguous tokens.

Ideas

• I1. Vowel-marked forms reduce ambiguity. Test it on tools/hbo or on he input: count analyses per token on the bare text and on a fully vowelled form, and report the drop. This measures how much an unpointed reading costs before any model is built.
• I2. Report coverage in two numbers: tokens without an analysis and types without an analysis. Types show the lexicon gaps more clearly (5.3 percent) than tokens (1.75 percent). A coverage gate on a held-out treebanks/he/ slice would show where the lexicon of he is thin.
• I3. Train a small tag disambiguator from a few fully checked files and report accuracy on ambiguous tokens only. The paper's 80 percent on ambiguous tokens, from 18 files, is a baseline for a Modern Hebrew disambiguator in 12. Mova and Modern Hebrew: where it stands, what it would need.

For Mova

I1 is a note for 2. Script and spelling and a test for tools/hbo in unpointed mode. I2 is a gate for treebanks/he/ and the he lexicon. I3 belongs in 11. Modern Hebrew in Universal Dependencies and the plan in 12. Mova and Modern Hebrew: where it stands, what it would need. The corpus is child speech and its licence is non-commercial, so only the measures and methods are used, not the data.

Cite

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

@inproceedings{albert2012morphologically,
  title = {{A Morphologically Annotated Hebrew CHILDES Corpus}},
  author = {Albert, Aviad and MacWhinney, Brian and Nir, Bracha and Wintner, Shuly},
  booktitle = {Proceedings of the Workshop on Computational Models of Language Acquisition and Loss},
  year = {2012},
  url = {https://aclanthology.org/W12-0904/}
}

References

6 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ruth A. Berman and Jürgen Weissenborn. Acquisition of word order: A crosslinguistic study. Final Report. German-Israel Foundation for Research and Development (GIF), 1991. find
2. Alon Itai and Shuly Wintner. Language resources for Hebrew. Language Resources and Evaluation, 42 (1):75–98, March 2008. find
3. Brian MacWhinney. The CHILDES system. American Journal of Speech Language Pathology, 5:5–14, 1996. find
4. Brian MacWhinney. The CHILDES Project: Tools for Analyzing Talk. Lawrence Erlbaum Associates, Mahwah, NJ, third edition, 2000. find
5. Bracha Nir, Brian MacWhinney, and Shuly Wintner. A morphologically-analyzed CHILDES corpus of Hebrew. In Proceedings of the Seventh conference on International Language Resources and Evaluation (LREC’10), pages 1487–1490, Valletta, Malta, May 2010. European Language Resources Association (ELRA). ISBN 2-9517408-6-7. find our card
6. Kenji Sagae, Eric Davis, Alon Lavie, Brian MacWhinney, and Shuly Wintner. Morphosyntactic annotation of CHILDES transcripts. Journal of Child Language, 37(3):705–729, 2010. doi: 10.1017/S0305000909990407. URL http://journals.cambridge.org/article_S0305000909990407. ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2012-albert-morphologically-annotated-hebrew-childes-corpus/