Elamur Research⌕
☰
home / papers / 2016 / Hypergraph Modelization of a Syntactically Annotated English Wikipedia Dump

Hypergraph Modelization of a Syntactically Annotated English Wikipedia Dump

Paper · 2016 · Edmundo Pavel Soriano Morales, Julien Ah-Pine and Sabine Loudcher

Summary

The paper presents SAEWD (Syntactically Annotated English Wikipedia Dump), a parsed snapshot of English Wikipedia that emphasizes syntax over semantics. Earlier Wikipedia parses had several gaps. They lacked constituency information, used awkward output formats, were sometimes no longer accessible, and did not ship the code that produced them. The authors took the November 2014 dump, about 4.7 million articles, and cleaned it with WikiExtractor. They then ran Stanford CoreNLP for sentence splitting, tokenization, lemmatization, PoS tagging, basic dependency parsing and constituency parsing. Each token is stored with its PoS tag, its dependency relation and its full constituency branch from the root down to the word, flattened into one line. The output reuses the column format of an earlier Wikipedia snapshot. As a proof of concept, they also built a hypergraph linguistic network from a subset of articles about NLP and Text Mining. In this network, words are linked through shared phrases, dependency relations and sentences. The corpus, the parsing tool with its source code, and the hypergraph are all released publicly, so graph-based and syntax-aware experiments can start without a preprocessing stage.

Key points

  • The parsed corpus contains 1,889,769,908 tokens, 8,761,691 types and 84,760,512 sentences, averaging 22.30 tokens per sentence.
  • The pipeline has three stages:
    • WikiExtractor removes tables, lists, references, markup and HTML, and splits the text into 429 folders of 100 files each, about 300 kB per file.
    • The Stanford CoreNLP left3words tagger assigns PoS tags.
    • The Stanford Shift-Reduce parser produces both the dependency and the constituency analyses.
  • Dependencies use the basic (uncollapsed) scheme with Universal Dependencies labels, so that every relation is kept.
  • Constituency paths give each chunk node a numeric identifier tied to the grammar rule that built it. For example, "brigand" gets the path NP22→S97.
  • The hypergraph is stored as a sparse incidence matrix in Matrix Market format, with JSON metadata. The authors flag possible redundancy from adding one hyperedge per noun phrase, which makes the number of columns grow quickly.

Interesting ideas

  • I1. Attaching to every token its flattened root-to-leaf constituency branch turns tree structure into a per-word feature that is cheap to read and compare.
  • I2. Chunk labels carry an ID that depends on the production rule. Two NPs built from different daughters therefore stay distinguishable, which keeps grammar-rule information that a bare label would lose.
  • I3. A word-by-hyperedge incidence matrix unifies several relation types: phrase membership, a dependency relation with its head, and sentence co-occurrence. One sparse structure can then be queried for many graph-based experiments.
  • I4. The authors prefer basic dependencies over collapsed ones so that no relation is merged away. Lossless syntax is kept as the stored layer, and any simplification can be derived later.
  • I5. The corpus is sharded into many small fixed-size files with article IDs and URLs kept as separators, and the tool is released alongside the data. This keeps memory use low, makes multithreading easy and lets others regenerate the resource.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tree): Storing a compact path string next to each token's UD head and deprel would let tools inspect and diff a token's structural position between witness parsers without rebuilding trees.
  • I2 (tree): Rule-specific chunk IDs could serve as features or diagnostics when converting between UD dialects or explaining why the parser grouped words a certain way.
  • I3 (world): A similar multi-relation incidence index over compiled corpora could quickly answer questions such as "which lemmas appear as nsubj of become" when building verb-class or ontology knowledge.
  • I4 (frames): Frame extraction should read from an uncollapsed UD layer and derive simplified views on demand, keeping each role assignment traceable to explicit arcs.
  • I5 (learn): Silver-data production could use small sharded files with source IDs and a published generation script, so that any silver tree can be reproduced and traced back to its article.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Université de LyonLyon · France4 works · 372 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksLiterature and Literary Theory · 519 works

🧭 Topics 3Research subjects and languages the work is about.English · 2,263 worksSyntax & grammar formalisms · 214 worksTreebanks · 126 works

Cite

Cite as Edmundo Pavel Soriano Morales, Julien Ah-Pine and Sabine Loudcher. 2016. Hypergraph Modelization of a Syntactically Annotated English Wikipedia Dump. Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16).

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

@inproceedings{morales2016hypergraph,
  title = {{Hypergraph Modelization of a Syntactically Annotated English Wikipedia Dump}},
  author = {Edmundo Pavel Soriano Morales, Julien Ah-Pine and Sabine Loudcher},
  booktitle = {Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)},
  year = {2016},
  doi = {10.63317/346oz8gtzhz6},
  url = {https://doi.org/10.63317/346oz8gtzhz6}
}

References

16 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Bergsma, S., Post, M., and Yarowsky, D. (2012). Stylometric Analysis of Scientific Articles. In Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL HLT ’12, pages 327– 337, Stroudsburg, PA, USA. Association for Computational Linguistics. find
  2. Charton, E. and Torres-Moreno, J.-M. (2010). NLGbAse: A Free Linguistic Resource for Natural Language Processing Systems. In Proceedings of the Seventh conference on International Language Resources and Evaluation (LREC’10), Valletta, Malta, May. European Language Resources Association (ELRA). find
  3. Choudhury, M. and Mukherjee, A. (2009). The structure and dynamics of linguistic networks. In Ganguly, N., Deutsch, A., and Mukherjee, A., editors, Dynamics On and Of Complex Networks, Modeling and Simulation in Science, Engineering and Technology, pages 145–166. Birkhäuser Boston. ↗
  4. Estrada, E. and Rodriguez-Velazquez, J. A. (2005). Complex networks as hypergraphs. arXiv preprint physics/0505137. find
  5. Flickinger, D., Oepen, S., and Ytrestøl, G. (2010). WikiWoods: Syntacto-Semantic Annotation for English Wikipedia. In Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC’10), Valletta, Malta, May. European Language Resources Association (ELRA).
  6. Giuseppe Attardi. (2015). WikiExtractor. https://github.com/attardi/wikiextractor. ↗
  7. Jordi Atserias, Hugo Zaragoza, M. C. and Attardi, G. (2008). Semantically Annotated Snapshot of the English Wikipedia. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC’08), Marrakech, Morocco, May. European Language Resources Association (ELRA).
  8. Katz, G., Shtock, A., Kurland, O., Shapira, B., and Rokach, L. (2014). Wikipedia-based query performance prediction. In Proceedings of the 37th international ACM SI-GIR conference on Research & development in information retrieval, pages 1235–1238. ACM. find
  9. Manning, C. D., Surdeanu, M., Bauer, J., Finkel, J., Bethard, S. J., and McClosky, D. (2014). The Stanford CoreNLP Natural Language Processing Toolkit. In Proceedings of 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 55–60.
  10. Massung, S., Zhai, C., and Hockenmaier, J. (2013). Structural parse tree features for text representation. In Semantic Computing (ICSC), 2013 IEEE Seventh International Conference on, pages 9–16. find
  11. Rada Mihalcea and Dragomir Radev. (2011). Graph-based Natural Language Processing and Information Retrieval. Cambridge University Press. Cambridge Books Online. ↗
  12. Sagae, K. and Gordon, A. S. (2009). Clustering Words by Syntactic Similarity Improves Dependency Parsing of Predicate-Argument Structures. In International Conference on Parsing Technologies (IWPT-09), Paris, France, October. find
  13. Schenkel, R., Suchanek, F. M., and Kasneci, G. (2007). YAWN: A semantically annotated wikipedia XML corpus. In Datenbanksysteme in Business, Technologie und Web (BTW 2007), 12. Fachtagung des GI-Fachbereichs ”Datenbanken und Informationssysteme” (DBIS), Proceedings, 7.-9. März 2007, Aachen, Germany, pages 277–291. find
  14. Shaoul, C. and Westbury, C. (2010). The Westbury Lab Wikipedia Corpus. http://www.psych. ualberta.ca/˜westburylab/downloads/ westburylab.wikicorp.download.html. ↗
  15. Vulić, I., De Smet, W., and Moens, M.-F. (2011). Identifying word translations from comparable corpora using latent topic models. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: short papers-Volume 2, pages 479–484. Association for Computational Linguistics. find
  16. Wu, F. and Weld, D. S. (2010). Open information extraction using wikipedia. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, ACL ’10, pages 118–127, Stroudsburg, PA, USA. Association for Computational Linguistics.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/morales-2016-hypergraph-modelization-syntactically-annotated-english/