[Mova](https://md.mova.elamur.ai/free/en/) · [this page in HTML](https://mova.elamur.ai/free/en/docs/papers/aclbulk-2025-shmidman-restoring-missing-spaces-scraped-hebrew/) [home](https://md.mova.elamur.ai/free/en/) / [docs](https://md.mova.elamur.ai/free/en/docs/) / [papers](https://md.mova.elamur.ai/free/en/docs/papers/) / [Modern Hebrew](https://md.mova.elamur.ai/free/en/docs/papers/he/) / Restoring Missing Spaces in Scraped Hebrew Social Media # Restoring Missing Spaces in Scraped Hebrew Social Media **Authors:** [Avi Shmidman](https://md.mova.elamur.ai/free/en/docs/papers/authors/avi-shmidman/); [Shaltiel Shmidman](https://md.mova.elamur.ai/free/en/docs/papers/authors/shaltiel-shmidman/) · **Year:** 2025 · **Venue:** Proceedings of the Tenth Workshop on Noisy and User-generated Text **Link:** [aclanthology.org/2025.wnut-1.3](https://aclanthology.org/2025.wnut-1.3/) · library `aclbulk-2025-shmidman-restoring-missing-spaces-scraped-hebrew` · **Licence:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) (ACL Anthology) [**0** citations](https://openalex.org/works?filter=cites:W4411118982) openalex · 2026-10-18[**1** citations in our library](https://md.mova.elamur.ai/free/en/docs/papers/cited-by/aclbulk-2025-shmidman-restoring-missing-spaces-scraped-hebrew/) of the 140673 papers whose bibliographies we read[Proceedings of the Tenth Workshop on Noisy and User-generated Text](https://aclanthology.org/volumes/2025.wnut-1/)[licence](https://creativecommons.org/licenses/by/4.0/) - [Publisher's page](https://aclanthology.org/2025.wnut-1.3/) - [PDF](https://aclanthology.org/2025.wnut-1.3.pdf) - [DOI 10.18653/v1/2025.wnut-1.3](https://doi.org/10.18653/v1/2025.wnut-1.3) - [OpenAlex](https://openalex.org/W4411118982) **Authors (2)** 1. [Avi Shmidman](https://openalex.org/A5013212409) h 5 · 122 citations · 53 works 2. [Shaltiel Shmidman](https://openalex.org/A5041361841) h 3 · 30 citations · 21 works Citations and author metrics from openalex (2026-10-18); authors with the same name are not always told apart. ## What it is about Scraped web text often runs two words together, and in unvocalised Hebrew one conflated string can usually be split in several valid ways, so a dictionary lookup does not settle it. The authors treat space restoration as character labelling (is there a space before this character?). They pretrain a character-level BERT (DictaBERT-char, the DictaBERT corpus and recipe, context 2048) and fine-tune it on 150,000 clean sentences in which 20% have 1 to 4 spaces removed. On 6,000 sentences from FineWeb2 with 1 to 4 spaces removed, the character model at threshold 0.9 reaches F1 0.968 (precision 97.0, recall 96.7). That is above GPT-4o (F1 0.961) and just below a 7B DictaLM 2.0 fine-tune (F1 0.976), and it runs far faster than the 7B model. The paper's error analysis lists three failure types: extra typos, proper nouns split into common words, and long words ending in a rare suffix. ## Ideas - I1. A post-filter on the character model's splits, from the paper's own failure list: accept a predicted split only if the second part is not a known Hebrew grammatical suffix sequence, so that a suffix is not cut off its word. - I2. A proper-noun guard: before splitting, look up the candidate pieces in a name list (or a PROPN tag from a parse) and keep names whole; the paper names this as the fix for its proper-noun errors. - I3. Measure a rule-based splitter on the released 6,000-sentence test set (a publicly released corpus, used here only as a measurement) and report precision and recall against the paper's 0.968 F1 at threshold 0.9, with the same setup on both sides. ## For Mova I1 and I2 are preprocessing rules for raw Hebrew text: they belong in `tools/hbo` before the segmenter, and I2 uses the lexicon notes in [8. Lexicon and register](https://md.mova.elamur.ai/free/en/docs/languages/he/08-lexicon-register/). I3 is a measurement for `tools/hbo`; the suffix list itself is a table from [Nouns, adjectives, pronouns and numerals](https://md.mova.elamur.ai/free/en/docs/languages/he/04-nouns-adjectives-pronouns/). ## Cite The paper itself, not this card. BibTeX from the library's record of the work. [DOI 10.18653/v1/2025.wnut-1.3](https://doi.org/10.18653/v1/2025.wnut-1.3) ``` @inproceedings{shmidman2025restoring, title = {{Restoring Missing Spaces in Scraped Hebrew Social Media}}, author = {Shmidman, Avi and Shmidman, Shaltiel}, booktitle = {Proceedings of the Tenth Workshop on Noisy and User-generated Text}, year = {2025}, doi = {10.18653/v1/2025.wnut-1.3}, url = {https://doi.org/10.18653/v1/2025.wnut-1.3} } ``` - [← previous OtoBERT: Identifying Suffixed Verbal Forms in Modern Hebrew Literature](https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2024-shmidman-otobert-identifying-suffixed-verbal-forms/) - [next → Towards Safer Hebrew Communication: A Dataset for Offensive Language Detoxification](https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2025-vanetik-safer-hebrew-communication-dataset-offensive/) - [↑ back to Modern Hebrew](https://md.mova.elamur.ai/free/en/docs/papers/he/) ## References 10 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Vitaly Shalumov and Harel Haskey. 2023. Hero: Roberta and longformer hebrew language models. arXiv:2304.11077. [↗](https://arxiv.org/abs/2304.11077) [our card](https://md.mova.elamur.ai/free/en/docs/papers/langstream-2023-shalumov-hero-roberta-longformer-hebrew-language/) 2. Avihay Chriqui and Inbal Yahav. 2021. Hebert & hebemo: a hebrew bert model and a tool for polarity analysis and emotion recognition. Preprint, arXiv:2102.01909. [↗](https://arxiv.org/abs/2102.01909) [our card](https://md.mova.elamur.ai/free/en/docs/papers/langstream-2021-chriqui-hebert-hebemo-hebrew-bert-model/) 3. Shaltiel Shmidman, Avi Shmidman, Amir DN Cohen, and Moshe Koppel. 2024a. Adapting llms to hebrew: Unveiling dictalm 2.0 with enhanced vocabulary and instruction capabilities. Preprint, arXiv:2407.07080. [↗](https://arxiv.org/abs/2407.07080) [our card](https://md.mova.elamur.ai/free/en/docs/papers/arxivrecent-2024-shmidman-adapting-llms-hebrew-unveiling-dictalm/) 4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805. [↗](https://arxiv.org/abs/1810.04805) 5. Shaltiel Shmidman, Avi Shmidman, Moshe Koppel, and Reut Tsarfaty. 2024b. MRL parsing without tears: The case of Hebrew. In Findings of the Association for Computational Linguistics ACL 2024, pages 4537–4550, Bangkok, Thailand and virtual meeting. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Shaltiel+Shmidman%2C+Avi+Shmidman%2C+Moshe+Koppel%2C+and+Reut+Tsarfaty%2E+2024b%2E+MRL+parsing+without+tears%3A+The+case+of+Hebrew%2E+In+Findings+of+the+Association+for+Computational+Linguistics+ACL+2024%2C+pages+453) [our card](https://md.mova.elamur.ai/free/en/docs/papers/findings-2024-shmidman-mrl-parsing-without-tears-case/) 6. Tau NLP Group. 2023. Tavbert: Hebrew character-based bert model. Accessed: 2025-01-21. [find](https://scholar.google.com/scholar?q=Tau+NLP+Group%2E+2023%2E+Tavbert%3A+Hebrew+character%2Dbased+bert+model%2E+Accessed%3A+2025%2D01%2D21%2E) 7. Eylon Gueta, Avi Shmidman, Shaltiel Shmidman, Cheyn Shmuel Shmidman, Joshua Guedalia, Moshe Koppel, Dan Bareket, Amit Seker, and Reut Tsarfaty. 2023. Large pre-trained models with extra-large vocabularies: A contrastive analysis of hebrew bert models and a new one to outperform them all. Preprint, arXiv:2211.15199. [↗](https://arxiv.org/abs/2211.15199) [our card](https://md.mova.elamur.ai/free/en/docs/papers/hbo-2022-gueta-large-pre-trained-models-extra/) 8. Reut Tsarfaty, Shoval Sadde, Stav Klein, and Amit Seker. 2019. What‘s wrong with Hebrew NLP? and how to make it right. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP): System Demonstrations, pages 259–264, Hong Kong, China. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Reut+Tsarfaty%2C+Shoval+Sadde%2C+Stav+Klein%2C+and+Amit+Seker%2E+2019%2E+What%E2%80%98s+wrong+with+Hebrew+NLP%3F+and+how+to+make+it+right%2E+In+Proceedings+of+the+2019+Conference+on+Empirical+Methods+in+Natural+Language+Pr) [our card](https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2019-tsarfaty-what-s-wrong-hebrew-nlp/) 9. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de Las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mT5: A massively multilingual pre-trained text-to-text transformer. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 483–498, Online. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Albert+Q%2E+Jiang%2C+Alexandre+Sablayrolles%2C+Arthur+Mensch%2C+Chris+Bamford%2C+Devendra+Singh+Chaplot%2C+Diego+de+Las+Casas%2C+Florian+Bressand%2C+Gianna+Lengyel%2C+Guillaume+Lample%2C+Lucile+Saulnier%2C+L%C3%A9lio+Renard+Lav) 10. Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Fei Wu, and Guoyin Wang. 2024. Instruction tuning for large language models: A survey. Preprint, arXiv:2308.10792. [↗](https://arxiv.org/abs/2308.10792)