Elamur Research⌕
☰
home / papers / 2014 / A Bayesian Mixed Effects Model of Literary Character

A Bayesian Mixed Effects Model of Literary Character

Summary

The paper addresses unsupervised discovery of character types ("personas") in fiction. Earlier models treated a persona as the sole source of all words linked to a character, so learned types tended to group characters by author or period instead of by role. Bamman, Underwood and Smith propose a hierarchical Bayesian mixed effects model in which a character's descriptive vocabulary is jointly explained by its latent persona and by extra-linguistic factors, here the author.

They apply it to 15,099 English novels (1700–1899) from the HathiTrust Digital Library. A scalable NLP pipeline extracts the actions, possessions and attributes tied to each character. The model was compared against preregistered similarity judgments by a literary scholar. The Author/Persona model was best at telling apart different character types within one author's work, while a Persona Regression alternative did better at grouping characters by author.

This matters because it gives humanists a flexible way to separate character from authorial style. The same approach could later factor out genre, period or narrative point of view.

Key points

  • Corpus: 469,200 English volumes were filtered to 32,209 prose-fiction volumes. After removing duplicates and merging multi-volume works, 15,099 narratives remained, totaling about 1.8 billion tokens.
  • Persona definition: A persona is a distribution over four dependency-based role types (agent, patient, possessive, predicative), extracted using Stanford typed dependencies.
  • Pipeline: The pipeline uses the Stanford POS tagger, the linear-time MaltParser and the Stanford NER, plus greedy name-variant clustering.
    • About 74% of character references are pronouns.
    • A pronominal coreference classifier was trained on 832 annotated links from 3 novels. It reached 82.7% accuracy in 10-fold cross-validation.
  • Vocabulary reduction: 100-dimensional skip-gram embeddings were learned for 512,344 word types. These were clustered via K-means into 1,000 groups with hierarchical bitstring codes.
  • Findings: Learned personas (e.g., with P = 50) are clearly gendered and show coherent trends over time, even though gender was never given to the model. Comparisons involving first-person narrators were a major source of error for all models.

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

Fields: Linguistics & languages

Cited by

99 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 93 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/p14-1035

@inproceedings{bamman2014bayesian,
  title = {{A Bayesian Mixed Effects Model of Literary Character}},
  author = {Bamman, David and Underwood, Ted and Smith, Noah A.},
  booktitle = {Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2014},
  doi = {10.3115/v1/p14-1035},
  url = {https://doi.org/10.3115/v1/p14-1035}
}

References

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

  1. Galen Andrew and Jianfeng Gao. 2007. Scalable training of l 1 -regularized log-linear models. In Proc. of ICML. find
  2. David Bamman, Brendan O’Connor, and Noah A. Smith. 2013. Learning latent personas of film characters. Proc. of ACL. find our card
  3. David Bamman, Ted Underwood, and Noah A. Smith. 2014. Appendix to ‘A Bayesian mixed effects model of literary character’. Technical report, Carnegie Mellon University, University of Illinois-Urbana Champaign. find
  4. David M. Blei, Andrew Y. Ng, and Michael I. Jordan. 2003. Latent Dirichlet allocation. Journal of Machine Learning Research, 3:993–1022. find our card
  5. Wayne Booth. 1961. The Rhetoric of Fiction. University of Chicago Press, Chicago. find
  6. Peter F. Brown, Peter V. deSouza, Robert L. Mercer, Vincent J. Della Pietra, and Jenifer C. Lai. 1992. Class-based n-gram models of natural language. Computational Linguistics, 18(4):467–479. find
  7. Nathanael Chambers. 2013. Event schema induction with a probabilistic entity-driven model. In Proc. of EMNLP, Seattle, Washington, USA. find
  8. Stanley F. Chen and Roni Rosenfeld. 2000. A survey of smoothing techniques for me models. IEEE Transactions on Speech and Audio Processing, 8(1):37–50. find
  9. Jackie Chi Kit Cheung, Hoifung Poon, and Lucy Vanderwende. 2013. Probabilistic frame induction. In Proc. of NAACL. find
  10. Peter T. Davis, David K. Elson, and Judith L. Klavans. 2003. Methods for precise named entity matching in digital collections. In Proc. of JCDL, Washington, DC, USA. find
  11. Marie-Catherine de Marneffe and Christopher D. Manning. 2008. Stanford typed dependencies manual. Technical report, Stanford University. find
  12. Greg Durrett, David Hall, and Dan Klein. 2013. Decentralized entity-level modeling for coreference resolution. In Proc. of ACL. find our card
  13. Jacob Eisenstein, Amr Ahmed, and Eric P. Xing. 2011. Sparse additive generative models of text. In Proc. of ICML. find
  14. David K. Elson, Nicholas Dames, and Kathleen R. McKeown. 2010. Extracting social networks from literary fiction. In Proc. of ACL, Stroudsburg, PA, USA. find
  15. Jenny Rose Finkel, Trond Grenager, and Christopher Manning. 2005. Incorporating non-local information into information extraction systems by gibbs sampling. In Proc. of ACL. find
  16. E. M. Forster. 1927. Aspects of the Novel. Harcourt, Brace & Co. find
  17. Joshua Goodman. 2001. Classes for fast maximum entropy training. In Proc. of ICASSP. find
  18. Joshua Goodman. 2004. Exponential priors for maximum entropy models. In Proc. of NAACL. find
  19. Aria Haghighi and Dan Klein. 2010. Coreference resolution in a modular, entity-centered model. In Proc. of NAACL. find our card
  20. Jun’ichi Kazama and Jun’ichi Tsujii. 2003. Evaluation and extension of maximum entropy models with inequality constraints. In Proc. of EMNLP. find
  21. Suzanne Keen. 2003. Narrative Form. Macmillan, Basingstoke. find
  22. Palgrave Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. Proc. of ICLR. find
  23. Frederic Morin and Yoshua Bengio. 2005. Hierarchical probabilistic neural network language model. In Proc. of AISTATS. find
  24. Joakim Nivre, Johan Hall, Jens Nilsson, Atanas Chanev, Gülsen Eryigit, Sandra Kübler, Svetoslav Marinov, and Erwin Marsi. 2007. Maltparser: A language-independent system for data-driven dependency parsing. Natural Language Engineering, 13:95–135, 5. find
  25. Vladimir Propp. 1998. Morphology of the Folktale. University of Texas Press, 2nd edition. find
  26. Roni Rosenfeld. 1996. A maximum entropy approach to adaptive statistical language modelling. Computer Speech and Language, 10(3):187 – 228. find
  27. Yanchuan Sim, Brice D. L. Acree, Justin H. Gross, and Noah A. Smith. 2013. Measuring ideological proportions in political speeches. In Proc. of EMNLP, Seattle, Washington, USA. find
  28. Kristina Toutanova, Dan Klein, Christopher D. Manning, and Yoram Singer. 2003. Feature-rich part-of-speech tagging with a cyclic dependency network. In Proc. of NAACL. find
  29. Ted Underwood, Michael L Black, Loretta Auvil, and Boris Capitanu. 2013. Mapping mutable genres in structurally complex volumes. In Proc. of IEEE International Conference on Big Data. find
  30. Alex Woloch. 2003. The One vs. the Many: Minor Characters and the Space of the Protagonist in the Novel. Princeton University Press, Princeton NJ. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/bamman-2014-bayesian-mixed-effects-model-literary/