Elamur Research⌕
☰
home / papers / 2012 / Dating medieval English charters

Dating medieval English charters

Paper · 2012 · Gelila Tilahun, Andrey Feuerverger and Michael Gervers

Summary

Most English property charters from the period between the Norman Conquest (1066) and about 1307 carry no date. Over a million survive, and fewer than 10% are dated, yet these documents are central evidence for the social and economic history of the era. Dating them by hand is slow, and it can compound mistakes, for example when similar personal names repeat across generations. The authors build statistical dating methods on the DEEDS corpus, about 10,000 Latin charters at the University of Toronto, each dated to within a year. The methods exploit how the use of words and word sequences ("shingles") changes over time. They compare three estimators: kNN averaging over the dates of the most similar dated documents; a new maximum prevalence (MP) method, which models the time-varying probability of each shingle and picks the date at which the document is most probable; and the minimum of a nonparametric quantile regression of document distances against known dates. Tuning uses cross-validation and kernel smoothing. MP performed best, which matters because cheaper automatic dating could open a huge undated archive to historians.

Key points

  • The data come from the DEEDS corpus, roughly 10,000 Latin charters from the tenth to early fourteenth centuries. Almost all originate in England or Wales, a small number come from Scotland, and every one has an issue date known to within a year.
  • Three estimators are compared: kNN with document distances, maximum prevalence (built on local log-likelihood and smoothing), and quantile regression minimisation. The kNN methods come from earlier work by Feuerverger et al., and MP is the new contribution.
  • MP won because it tracks each shingle's behaviour over time. It automatically down-weights very frequent shingles, whose usage is nearly constant across time. It does not, however, discount very rare ones.
  • Single-word shingles came surprisingly close to the best results, and two-word shingles were only modestly better. Combining shingle sizes or blending methods gave no striking gains.
  • In a real test, MP was trained on 2608 documents and applied to a newly found charter from Brock University. It dated the charter to 1246, while historians placed it at about 1235–1245, most likely 1238–1242. The project's target accuracy is ±3 years, 95% of the time.

Interesting ideas

  • I1. MP treats the date as a hidden parameter: it smooths each feature's usage curve over time and picks the time that best explains the whole document, so features that never change over time stop mattering without needing a stop-list.
  • I2. The authors argue that distances between documents should depend on the collection they come from, weighting each shared pattern by its length, how long it stays in use, and how widespread it is across the corpus, rather than counting overlaps in absolute terms.
  • I3. Instead of averaging the dates of the nearest neighbours, the third method fits a lower quantile curve of distance against date and reads off where it bottoms out, which is a robust way to locate where a query best matches a reference set.
  • I4. The authors validate their best method on one genuinely new document that historians had dated independently, alongside the cross-validation results.
  • I5. The authors are exploring how to order a collection in time, up to reversal, without using any known dates, relying only on gradual drift in features.

For Mova

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

  • I1 (world): time-indexed feature curves could estimate the period or register of a text and flag anachronistic wording, providing a traceable "when was this written" fact for the text's world.
  • I2 (learn): when choosing silver sentences or comparing parser outputs, agreement on corpus-rare structures could count more than agreement on ubiquitous ones.
  • I3 (learn): the same curve-minimum trick could locate which slice of training data (domain, genre, time) a failing input most resembles, helping target repair rules.
  • I4 (learn): a small set of truly unseen, expert-annotated texts kept outside all tuning loops would act as an honest external check next to the negative controls.
  • I5 (world): a similar ordering of a story's passages by how its entities and states shift could serve as a consistency check on the event order the system extracts.

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

🎓University of TorontoToronto · Canada110 works · 179,061 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksGeneral Social Sciences · 254 worksLiterature and Literary Theory · 541 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,379 works

Cite

Cite as Gelila Tilahun, Andrey Feuerverger and Michael Gervers. 2012. Dating medieval English charters. arXiv:1301.2405 · Annals of Applied Statistics 2012, Vol. 6, No. 4, 1615-1640.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1214/12-aoas566 · arXiv 1301.2405

@misc{tilahun2012dating,
  title = {{Dating medieval English charters}},
  author = {Gelila Tilahun, Andrey Feuerverger and Michael Gervers},
  year = {2012},
  doi = {10.1214/12-aoas566},
  eprint = {1301.2405},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.1214/12-aoas566}
}

References

29 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Berry, M. W. and Browne, M. (2005). Understanding Search Engines—Mathematical Modeling and Text Retrieval, 2nd ed. SIAM, Philadelphia. find
  2. Broder, A. Z. (1998). On the resemblance and containment of documents. In International Conference on Compression and Complexity of Sequences (SEQUENCES’97), June 11–13 1997, Positano, Italy 21–29. IEEE Comput. Soc., Los Alamitos, CA. find
  3. de Jong, F., Rode, H. and Hiemstra, D. (2005). Temporal language models for the disclosure of historical text. In Proc. 16th Int. Conf. of the Assoc. for History and Computing 161–168. KNAW, Amsterdam. find
  4. Djeraba, C. (2003). Multimedia Mining—A Highway to Intelligent Multimedia Documents. Kluwer, Boston. find
  5. Domingos, P. and Pazzani, M. (1996). Beyond independence: Conditions for optimality of the Bayes classifier. In Proceedings of the 13th International Conference on Machine Learning 105–112. Association for Computing Machinery, New York. find
  6. Fan, J. and Gijbels, I. (2000). Local polynomial fitting. In Smoothing and Regression: Approaches, Computation, and Application (M. G. Schimek, ed.) 229–276. Wiley, New York. ↗
  7. Feuerverger, A., He, Y. and Khatri, S. (2012). Statistical significance of the Netflix challenge. Statist. Sci. 27 202–231. ↗
  8. Feuerverger, A., Hall, P., Tilahun, G. and Gervers, M. (2005). Distance measures and smoothing methodology for imputing features of documents. J. Comput. Graph. Statist. 14 255–262. MR2160812 ↗
  9. Feuerverger, A., Hall, P., Tilahun, G. and Gervers, M. (2008). Using statistical smoothing to date medieval manuscripts. In Beyond Parametrics in Interdisciplinary Research: Festschrift in Honor of Professor Pranab K. Sen (N. Balakrishnan, DATING MEDIEVAL ENGLISH CHARTERS E. Pena, M. J. Silvapulle, eds.). Inst. Math. Stat. Collect. 1 321–331. Inst. Math. Statist., Beachwood, OH. MR2462216 ↗
  10. Fiallos, R. (2000). An overview of the process of dating undated medieval charters: Latest results and future developments. In Dating Undated Medieval Charters (M. Gervers, ed.). Boydell Press, Woodbridge. find
  11. Gervers, M. (2000). Dating Undated Medieval Charters. Boydell Press, Woodbridge. find
  12. Gervers, M. and Hamonic, N. (2010). Pro amore dei : Diplomatic evidence of social conflict during the reign of King John. Preprint. find
  13. Hastie, T., Tibshirani, R. and Friedman, J. (2009). The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd ed. Springer, New York. MR2722294 find
  14. Kanhabua, N. and Norvag, K. (2008). Improving Temporal Language Models for Determining Time of Non-Timestamped Documents. Lecture Notes in Computer Science 5173. Springer, Berlin. find
  15. Kanhabua, N. and Norvag, K. (2009). Using Temporal Language Models for Documents Dating. Lecture Notes in Computer Science 5782. Springer, Berlin. find
  16. Koenker, R. (2005). Quantile Regression. Econometric Society Monographs 38. Cambridge Univ. Press, Cambridge. MR2268657 find
  17. Loader, C. (1999). Local Regression and Likelihood. Springer, New York. MR1704236 ↗
  18. Luhn, H. P. (1958). The automatic creation of literature abstracts. IBM J. Res. Develop. 2 159–165. MR0090905 ↗
  19. Manning, C., Raghavan, P. and Schütze, H. (2008). Introduction to Information Retrieval. Cambridge Univ. Press, New York. ↗
  20. McGill, M., Koll, M. and Noreault, T. (1979). An evaluation of factors affecting document ranking by information retrieval systems. Technical Report. School of Information Studies, Syracuse Univ., Syracuse, NY. find
  21. Mosteller, F. and Wallace, D. (1963). Inference in an authorship problem. J. Amer. Statist. Assoc. 58 275–302. ↗
  22. Nadaraya, E. A. (1964). On estimating regression. Theory Probab. Appl. 10 186–190. ↗
  23. Quang, P. X., James, B., James, K. L. and Levina, L. (1999). Document similarity measure for the vector space model in information retrieval. NSASAG Problem 99-5. find
  24. Salton, G., Wang, A. and Yang, C. (1975). A vector space model for information retrieval. J. Amer. Soc. Inf. Sci. 18 613–620. find
  25. Simonoff, J. S. (1996). Smoothing Methods in Statistics. Springer, New York. MR1391963 ↗
  26. Tan, P. N., Steinbach, M. and Kumar, V. (2005). Introduction to Data Mining. Addison-Wesley, Reading. find
  27. Wand, M. P. and Jones, M. C. (1995). Kernel Smoothing. Monographs on Statistics and Applied Probability 60. Chapman & Hall, London. MR1319818 find
  28. Watson, G. S. (1964). Smooth regression analysis. Sankhyā Ser. A 26 359–372. MR0185765 find
  29. Zhang, J. and Korfhagen, R. (1999). A distance and angle similarity measure method. J. Amer. Soc. Inf. Sci. 50 772–778. G. Tilahun A. Feuerverger Department of Statistics University of Toronto 100 St. George Street Toronto, Ontario M5S 3G3 Canada E-mail: gelila@utstat.toronto.edu andrey@utstat.toronto.edu M. Gervers Department of History University of Toronto 100 St. George Street Toronto, Ontario M5S 3G3 Canada E-mail: m.gervers@utoronto.ca ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tilahun-2013-dating-medieval-english-charters/