Unsupervised Decomposition of a Document into Authorial Components

What it is about

The question is whether the authorial layers of a composite text can be separated without any labelled example. The test bed is the Hebrew Bible. Two books (Jeremiah and Ezekiel, and later Isaiah, Job and Proverbs) are "munged" into one artificial document, and the method must recover the two books. Clustering chapters by bag-of-words gives a normalised majority diagonal (NMD) of 51.5%, close to the chance floor of 50%. Clustering by synonym choice, where synsets come from words that the King James translation renders identically, gives 91 of 100 chapters on the right side. A two-stage method uses the reliable core of those clusters as training data for an SVM on generic words, and reaches 99 of 100. On the munged pairs most verses are assigned correctly, and the split agrees with the Priestly and non-Priestly consensus for over 90% of Pentateuch verses where a consensus exists.

Ideas

• I1. Synonym choice without a thesaurus: a synset is a set of source words that one translation renders with the same English word, with the concordance giving the senses. For Mova, the lemmas and Strong's numbers in macula dict could give Hebrew synsets, and the test is whether synonym choice separates the layers of the Pentateuch in The lexicon.
• I2. A two-stage design as a general method: a high-precision partial clustering from a sparse, rule-like feature set, then a small classifier on a denser feature set trained only on the cores. This fits the rule-first design of Mova, since the sparse stage can be a set of rules and the dense stage a small model that can be inspected.
• I3. A negative control: munge two books of known authorship and check that the method recovers them, with the naive 51.5% as the floor that must be beaten. A test that cannot show failure is not a gate; the munged set is for tests only and is not part of any release.

For Mova

I1 goes into The lexicon and tools/macula. I2 is a design note for world, where clusters of text can be built on a rule-first basis. I3 is a test under tools/hbo or tools/macula, with a dated run.

Cite

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

@inproceedings{koppel2011unsupervised,
  title = {{Unsupervised Decomposition of a Document into Authorial Components}},
  author = {Koppel, Moshe and Akiva, Navot and Dershowitz, Idan and Dershowitz, Nachum},
  booktitle = {Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies},
  year = {2011},
  url = {https://aclanthology.org/P11-1136/}
}

References

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

1. J. Astruc. 1753. Conjectures sur les mémoires originaux dont il paroit que Moyse s’est servi pour composer le livre de la Genèse. Brussels. find
2. R. E. Bee. 1971. Statistical methods in the study of the Masoretic text of the Old Testament. J. of the Royal Statistical Society, 134(1):611-622. find
3. M. J. Berryman, A. Allison, and D. Abbott. 2003. Statistical techniques for text classification based on word recurrence intervals. Fluctuation and Noise Letters, 3(1):L1-L10. find
4. J. E. Carpenter, G. Hartford-Battersby. 1900. The Hexateuch: According to the Revised Version. London. find
5. J. Clark and C. Hannon. 2007. A classifier system for author recognition using synonym-based features. Proc. Sixth Mexican International Conference on Artificial Intelligence, Lecture Notes in Artificial Intelligence, vol. 4827, pp. 839-849. find
6. I. S. Dhillon, Y. Guan, and B. Kulis. 2004. Kernel kmeans: spectral clustering and normalized cuts. Proc. ACM International Conference on Knowledge Discovery and Data Mining (KDD), pp. 551-556. find
7. S. R. Driver. 1909. An Introduction to the Literature of the Old Testament (8th ed.). Clark, Edinburgh. find
8. N. Graham, G. Hirst, and B. Marthi. 2005. Segmenting documents by stylistic character. Natural Language Engineering, 11(4):397-415. find
9. D. Guthrie, L. Guthrie, and Y. Wilks. 2008. An unsupervised probabilistic approach for the detection of outliers in corpora. Proc. Sixth International Language Resources and Evaluation (LREC'08), pp. 28-30. find
10. D. Holmes. 1994. Authorship attribution, Computers and the Humanities, 28(2):87-106. find
11. P. Juola. 2008. Author Attribution. Series title: Foundations and Trends in Information Retrieval . Now Publishing, Delft. find
12. M. Koppel, N. Akiva, and I. Dagan. 2006. Feature instability as a criterion for selecting potential style markers. J. of the American Society for Information Science and Technology, 57(11):1519-1525. find
13. M. Koppel, J. Schler, and S. Argamon. 2009. Computational methods in authorship attribution. J. of the American Society for Information Science and Technology, 60(1):9-26. find
14. D. L. Mealand. 1995. Correspondence analysis of Luke. Lit. Linguist Computing, 10(3):171-182. find
15. S. Meyer zu Eisen and B. Stein. 2006. Intrinsic plagiarism detection. Proc. European Conference on Information Retrieval (ECIR 2006), Lecture Notes in Computer Science, vol. 3936, pp. 565–569. find
16. K. Nigam, A. K. McCallum, S. Thrun, and T. M. Mitchell. 2000. Text classification from labeled and unlabeled documents using EM, Machine Learning, 39(2/3):103-134. find
17. M. H. Pope. 1965. Job (The Anchor Bible, Vol. XV). Doubleday, New York, NY. find
18. M. H. Pope. 1952. Isaiah 34 in relation to Isaiah 35, 40-66. Journal of Biblical Literature, 71(4):235-243. find
19. Y. Radday. 1970. Isaiah and the computer: A preliminary report, Computers and the Humanities, 5(2):65-73. find
20. E. Stamatatos. 2009. A survey of modern authorship attribution methods. J. of the American Society for Information Science and Technology, 60(3):538-556. find
21. J. Strong. 1890. The Exhaustive Concordance of the Bible. Nashville, TN. (Online edition: http://www.htmlbible.com/sacrednamebiblecom/kjvs trongs/STRINDEX.htm; accessed 14 November 2010.) ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2011-koppel-unsupervised-decomposition-document-into-authorial/